RESEARCH DOSSIER
Extinction, Inhibitory Learning, and Renewal
Abstract. Extinction is an experience-dependent decrease in acquired behavior after the removal or omission of an expected outcome. This fundamental research monograph synthesizes historical experiments, contemporary associative and computational theories, animal and human neuroscience, methodological research, and empirical syntheses to explain why diminished responding can later recover. It examines acquisition–expression distinctions, the formation and retrieval of inhibitory learning, spontaneous recovery, ABA/ABC/AAB renewal, reinstatement, reacquisition, latent-cause inference, reconsolidation hypotheses, Pavlovian and instrumental extinction, and context-sensitive neural systems. Recent human intracranial evidence, systematic reviews and meta-analyses, and preregistered replication failures constrain claims about mechanistic certainty and clinical translation. The conclusions distinguish experimental extinction from patient-level treatment efficacy, identify reproducible effects and their contextual moderators, and set out theory-discriminating research questions. This is a critical narrative synthesis of existing research, not a new experimental study, systematic review, or meta-analysis.
Keywords: extinction; inhibitory learning; renewal; ABA renewal; ABC renewal; context-dependent retrieval; Pavlovian conditioning; instrumental extinction; spontaneous recovery; reinstatement; rapid reacquisition; prediction error; latent cause inference; fear extinction; reconsolidation; exposure therapy; amygdala; hippocampus; ventromedial prefrontal cortex; replication; Artificial Era
Extinction as an Object of Psychological Science
Extinction is the experience-dependent decline in an acquired response when the contingency that once supported it changes. In Pavlovian conditioning, a conditioned stimulus that predicted an unconditioned event is presented without that event; in instrumental learning, an action no longer produces its established consequence. The result may be a reduction in freezing, autonomic reactivity, expectancy, approach, lever pressing, or avoidance, depending on the preparation. A complete scientific definition specifies the acquisition history, event omission, testing context, and response measure. The influential analysis in Bouton (2004) explains why the decline itself cannot identify whether an original association has been weakened, inhibited, or rendered less accessible at retrieval.
Extinction is simultaneously a behavioral phenomenon and an inference problem. A response is observed, while the acquired representation and its retrieval conditions are inferred from experiments that move beyond the immediate extinction curve. A decline can coexist with later spontaneous recovery, context-related renewal, or reinstatement after unsignaled outcomes. These recovery phenomena are especially informative because they demonstrate that an extinguished performance can remain sensitive to earlier learning. The experimental syntheses in Rescorla (2004a) and Vervliet et al. (2013) give extinction its comparative scope across appetitive, aversive, human, and animal procedures.
This monograph investigates what changes during extinction, what conditions govern expression of the changed behavior, and why response recovery follows an apparently successful learning episode. The central objects are inhibitory acquisition, contextual retrieval, temporal change, reinforcer expectancy, competing behavioral control, and the evidence required to discriminate these accounts. The investigation integrates Pavlovian and instrumental traditions while assigning clinical exposure, addiction, and neuroscience their appropriate evidential roles. The intended result is an explanatory map of established findings, rival mechanisms, methodological limits, and testable questions, rather than a new empirical dataset or a claim of therapeutic efficacy Todd et al. (2014).
Acquisition, Extinction, Retention, and Performance
An acquisition session establishes a behavioral history under specified event relations, but a measured acquisition curve is only one expression of that history. After extinction, a weak response may reflect new learning, a changed motivational state, a temporary response deficit, fatigue, or a shift in attention. Retention is assessed by delayed testing, ideally with controls for test-induced learning; transfer is assessed by changing context, stimulus, temporal interval, or response requirement. The separation between acquired control and momentary performance is foundational to interpretations of relapse Bouton (2002).
An extinction trial is an informative event when the organism expects an outcome and that outcome is omitted. In an aversive discrimination task, a conditioned danger cue may previously have been reinforced while a comparison cue remained unreinforced; during extinction, omission occurs following the formerly threatening cue. The resulting learning can affect explicit expectancy sooner than physiological response, or the measures can change in the opposite temporal order. Extinction strength therefore depends on what the organism predicted and on which component of the reaction the experiment measures Craske et al. (2014).
The distinction is sharpened by the practical difference between within-session reduction and later retrieval. Low responding at the last extinction trial does not establish that the same response will remain low after a day or in another setting. A robust design includes a baseline acquisition test, a specified extinction criterion, delayed and immediate tests, and comparison groups matched for relevant exposures. Recovery measures should be interpreted relative to performance just before testing, because differences in the starting level can mimic differential relapse Lonsdorf et al. (2017).
Origins of Experimental Extinction
Early experimental learning traditions treated the disappearance of a conditioned response as a problem of changing stimulus control. Pavlovian laboratory procedures made the contingency between cue and biologically significant event manipulable; instrumental laboratories examined behavior after reinforcement was withheld. By the later twentieth century, experimental questions had become more precise: Does omission abolish previous learning, or does the new training create additional information? What features of the environment signal that the earlier or later relation is relevant? These questions led from observable response decrement to competing models of associative representation Bouton (2004).
The work of Bouton and Bolles was decisive because experimental context was explicitly manipulated. In their 1979 study of contextual control, rats learned a conditioned suppression relation in one chamber and received nonreinforced cue presentations in another. Suppression returned when the cue was tested in the acquisition chamber, and renewal could also occur in a third context. This demonstration showed that an extinguished response could recover even after extensive exposure to the cue without the outcome. The published experimental design, rather than a retrospective slogan, defines the historical contribution Bouton & Bolles (1979a).
Research on reinstatement supplied a second historical line. Bouton and Bolles examined the return of conditioned fear after presentations of the aversive unconditioned event that were independent of the cue. Their results implicated contextual associations at the time of the unsignaled events and subsequent testing Bouton & Bolles (1979b). Bouton and King later used independent assessments of the context's associative value to distinguish contextual modulation of renewal from contextual excitation that contributes to reinstatement Bouton & King (1983). These studies established that superficially similar behavioral recoveries need distinct experimental explanations.
Contingency, Prediction, and the Informational Value of Omission
The core information in an extinction trial is a discrepancy between expected and observed outcomes. If a cue previously predicted shock, food, or another consequence, an omission under comparable circumstances can alter the organism's expectation. The change is clearest when the cue reliably activates the expectation and when the outcome would otherwise have been likely. Trial count alone is an incomplete dose measure: identical numbers of nonreinforced cue presentations can carry different informational value depending on prior reinforcement frequency, discriminative stimuli, and retrieval conditions. Inhibitory-learning applications emphasize expectancy violation for this reason Craske et al. (2014).
Prediction-error theories offer tractable descriptions of this update. A simple rule reduces an excitatory cue value when expected reinforcement is omitted; different learning rates may govern cues, contexts, and outcomes. Such models capture some aggregate curves but do not automatically explain why an old response recovers after a context switch. A model that stores only one context-free association will ordinarily predict monotonic replacement rather than history-sensitive retrieval. Explicit context representations, latent causes, or multiple associations are required to generate a wider class of extinction and renewal phenomena Gershman et al. (2013).
Contingency also differs from mere temporal proximity. The organism may discriminate a cue's conditional probability of reinforcement from the ambient outcome rate and use temporal structure as information. In human experiments, verbal expectancies and demand characteristics introduce an additional layer because a participant can form propositions about laboratory rules. The scientific issue is not which vocabulary is preferred but which experimental manipulation produces diverging predictions among candidate models. Controlled changes in cue, outcome, context, and delay are therefore indispensable to any claim that extinction taught a particular relationship Lonsdorf et al. (2017).
The Architecture of Evidence
Extinction research spans individual-animal experiments, human conditioning studies, clinical interventions, computational simulations, and neural recordings. The design with the strongest causal leverage for a particular mechanism is one in which that mechanism generates a distinctive prediction under manipulation. Randomly assigning contexts can test contextual retrieval; separately manipulating unconditioned-event presentations can test reinstatement; changing outcome identity can probe what has been learned. Behavioral replications across organisms offer convergence, while neural correlations supply a different level of explanation. None of these alone substitutes for the others Todd et al. (2014).
Systematic reviews and meta-analyses extend interpretation beyond single laboratories. The human renewal review by Wang et al. (2024) searched eighty studies and quantitatively pooled twenty-three, finding that conventional ABA tests tended to produce stronger renewal than some alternative designs. The separate multilevel meta-analysis by Bustamante et al. (2024) assembled twenty-five studies and thirty-seven critical comparisons of extinction conducted across multiple contexts. These results are valuable for assessing generality, yet their procedural diversity means that the exact prediction for an individual person or clinical treatment cannot be read directly from a pooled laboratory effect.
Methodological quality is a property of the link between claim and design. Critical questions include whether cue assignment was counterbalanced, whether participants who failed acquisition were excluded under preregistered rules, whether physiological artifacts were handled consistently, whether counterfactual outcomes were controlled, and whether tests used sufficiently few trials to prevent rapid re-extinction. The methodological framework in Lonsdorf et al. (2017) documents substantial researcher flexibility in human fear conditioning. Accordingly, this narrative synthesis presents original evidence, structured comparative interpretation, and clearly labeled uncertainties without claiming the protocol, exhaustive search, or effect-size synthesis of a new systematic review.
The Time Course of Extinction
Across nonreinforced trials, an acquired response often decreases, but that decline contains more than one process. Early trials may produce strong prediction errors, orienting, novelty, or frustration; later trials can bring expectation revision and changed response selection. The trial-by-trial slope is sensitive to reinforcement history, temporal spacing, initial performance, trial duration, and how response magnitude is scaled. A good analysis separates a change in average response from a change in its variability, latency, or stimulus specificity. The possibility of rapid within-session decline with weak delayed retention is one reason that terminal extinction performance cannot serve as the sole criterion of durable learning Bouton (2004).
Training amount creates an additional difficulty. Extensive nonreinforcement may provide many opportunities to learn that the outcome has stopped, but it also provides repeated exposure to the extinction environment, possibly strengthening the retrieval control of that environment. An animal can therefore show nearly complete extinction in a chamber while responding again when tested elsewhere. The effect is intelligible once the changed contingency and the conditions for recalling that contingency are analyzed separately. Such separation remains central to interpretations of emotional exposure and appetitive habit change Bouton (2002).
Temporal spacing itself modifies the learning situation. Massed omissions may be experienced as one sustained episode, whereas spaced presentations sample different internal and external states. Some schedules facilitate short-term decreases but yield less reliable retention, and a schedule optimal for one outcome may not generalize to other procedures. The literature on temporal recovery demonstrates that the acquisition-to-extinction interval matters as well as the interval from extinction to test. In four animal experiments, Rescorla (2004b) found greater spontaneous recovery when conditioning and extinction were closer together, challenging simple accounts that recovery merely reflects passive fading of extinction memory.
Spontaneous Recovery and Temporal Context
Spontaneous recovery describes renewed expression of an extinguished response after the passage of time without renewed conditioning. A delayed test is therefore a critical probe of persistence. The response can reappear after a procedure that produced little or no expression at the end of extinction; that difference indicates a change in retrieval conditions or competition between learning histories. A temporal-context account treats elapsed time as part of the background that changes between extinction and test, allowing earlier acquisition control to become more influential. This family of accounts is supported by a broad range of observations but must accommodate more than the passage of time alone Rescorla (2004a).
The design must separate time-dependent recovery from test exposure, circadian variation, maturation, and different opportunities to express the behavior. If one group is tested after a long interval and another immediately, procedural equivalence requires more than equal numbers of extinction trials. Testing itself usually adds additional cue-alone presentations and thereby creates new extinction experience. Early test trials are most diagnostic of retrieval, although they may be noisier. Delayed response elevation should be compared against a stable acquisition history and carefully measured terminal extinction responding Lonsdorf et al. (2017).
Rescorla's review Rescorla (2004a) cautioned against treating the mechanism of spontaneous recovery as singular. Different experimental preparations vary in their baseline conditioning, response topography, and susceptibility to competition or time-dependent changes. The specific finding that recovery can vary inversely with the acquisition-to-extinction interval Rescorla (2004b) also shows that the history before extinction influences the delayed outcome. Scientific claims about the decay of safety information should therefore identify the relevant timing and population, and cannot infer molecular memory decay from the mere presence of behavioral recovery.
Renewal Across Contexts
Renewal is the increase of an extinguished response after a change in contextual conditions between extinction and testing. The standard ABA procedure acquires responding in context A, extinguishes it in B, and tests it back in A. ABC renewal instead tests in a third context C, while AAB renewal places extinction in the original acquisition context and tests in a different one. These arrangements discriminate return to a training environment from the broader effect of testing outside the extinction environment. The original rat experiments Bouton & Bolles (1979a) established ABA and third-context renewal, while later studies expanded the range of controlled contextual arrangements.
Context comprises the configuration of external surroundings, interoceptive state, temporal circumstances, and task instructions capable of modulating which learning history controls behavior. Physical locations are one operational method, not the complete concept. Laboratory manipulations include changes in chamber properties, ambient light, screen background, timing, and occasion-setting cues. In a human differential-conditioning experiment Vansteenwegen et al. (2005), a manipulation of illumination between acquisition, extinction, and test produced renewal. The design makes the finding especially instructive: contextual control can arise with subtle contextual variations rather than requiring dramatic relocation or a clinical environment.
A systematic review and meta-analysis Wang et al. (2024) compared human renewal preparations and found stronger evidence for the conventional ABA arrangement than for ABC or ABBA variants, with meaningful heterogeneity in how studies defined and measured their contexts. It also identified evidence that multi-context extinction and greater similarity between acquisition and extinction contexts can reduce renewal. These patterns support context-sensitive retrieval as a useful organizing explanation, while avoiding a universal claim that all context shifts produce the same magnitude or even direction of recovery. Context functions are established by experimental relations rather than the names of experimental rooms.
Reinstatement After Unsignaled Outcomes
Reinstatement occurs when exposure to the unconditioned outcome after extinction increases responding to the previously extinguished cue, even without new cue–outcome pairings. In aversive conditioning, a few unsignaled aversive events may revive a defensive response; in appetitive procedures, reinforcer presentations can alter the motivational and contextual setting for an extinguished cue. The causal interpretation depends on whether the outcomes are presented in the same context later used for testing, whether general arousal changes, and whether the cue has been indirectly reconditioned Bouton & Bolles (1979b).
Bouton and Bolles Bouton & Bolles (1979b) reported experiments in which contextual stimuli helped mediate the effect of unsignaled outcomes. Bouton and King Bouton & King (1983) found that reinstatement was associated with measurable contextual excitation at test, whereas renewal could occur without analogous independently detected excitation or inhibition of the context. This difference matters because reinstatement is sometimes described loosely as return of the original learning, even though intervening outcomes can change the test environment itself. A proper reinstatement control evaluates the context and the unconditioned-event history rather than attributing all changes to the target cue.
In human fear-conditioning experiments, ethical constraints require carefully bounded aversive stimuli and consent, and the outcome may vary from an unpleasant sound to mild electrical stimulation. Reinstatement tests can be contaminated by changes in general distress or vigilance; outcome expectancy, differential skin conductance, and response specificity can help separate interpretations. The review of human extinction methods by Lonsdorf et al. (2017) emphasizes that variations in stimulus intensity, reinforcement schedule, scoring window, and exclusion procedures alter what is measured. The laboratory phenomenon supplies mechanistic hypotheses for clinical symptom recurrence, but does not by itself quantify relapse rates in diagnosed populations.
Rapid Reacquisition and the Survival of Prior Learning
Rapid reacquisition describes faster recovery of conditioned behavior when reinforcement resumes after extinction than during the first acquisition phase. It is often interpreted as evidence that earlier experience persists in some form. Yet faster responding also depends on the experimental comparison: the organism is older, familiar with the apparatus, and may have learned the response requirements. A decisive design matches trial structure and evaluates appropriate controls rather than concluding that a specific excitatory association survived solely from an early response increase Bouton (2004).
Extinction can preserve information about the original contingency while changing which information controls behavior. Renewal and reinstatement are particularly clear behavioral expressions of this possibility, and faster reacquisition may add convergent evidence when auxiliary learning is controlled. A multiple-memory interpretation permits both the initial expectation and later omission learning to contribute to action with weights determined by context, time, and current reinforcement. The resulting account can describe relapse as re-expression of a previously acquired behavioral policy rather than necessarily a wholly new episode of learning Todd et al. (2014).
An important unresolved issue is how to differentiate savings of the original association from enhanced attention, reinforcement sensitivity, or procedural learning. Crossed designs that manipulate the original outcome, the reinforcer identity, and context at reacquisition can refine the interpretation. Distinct dependent measures may yield different conclusions about what survived. The strongest claim is therefore conditional: recovery phenomena place substantial pressure on single-process erasure accounts, but an observed recovery does not identify a unique memory substrate or imply that extinction never changes the original representation Dunsmoor et al. (2015).
Inhibitory Learning and Competing Associations
Inhibitory-learning accounts hold that nonreinforcement generates information capable of suppressing the behavioral influence of earlier cue–outcome learning. A cue that once signaled an outcome can acquire a context-sensitive relation indicating that the expected outcome is absent. The word inhibition therefore refers to an experimentally inferred control of expression, not necessarily an anatomical inhibitory synapse or a general inability to respond. In some formulations, extinction establishes an inhibitory association; in others, the extinction episode becomes an alternative memory whose retrieval competes with acquisition. The analysis of Bouton (2004) evaluates these conceptual possibilities against recovery and transfer experiments.
A central prediction is that extinction memory may be more dependent on context than original acquisition, especially because nonreinforcement makes the cue ambiguous: the same cue has had different consequences on different occasions. Context can then select the relation that is currently appropriate. Thus returning to acquisition conditions after extinction increases the likelihood of retrieving the earlier relation, and moving to a third context can have the same effect. Behavioral evidence from Bouton & King (1983) suggests this selection does not require an independently measurable inhibitory associative value of the context itself.
Inhibitory learning is often invoked as an organizing principle for exposure procedures, but its experimental meaning must remain precise. A reduction in reported fear does not prove a newly acquired inhibitory association; expectancy updating, attentional disengagement, response regulation, and memory competition may also change. Mechanistic claims are stronger when outcome expectancy is probed, contextual transfer is tested, and later recovery is measured under controlled conditions. Craske et al. (2014) argues that exposure should emphasize learning about safety and violated expectancies while recognizing the need to evaluate durable change rather than only immediate fear reduction.
Context as an Occasion Setter and Retrieval Signal
A context need not directly predict a reinforcer in order to determine which cue–outcome relation governs behavior. An occasion setter changes the relevance of another stimulus's relationship to an outcome, making the effect conditional on the compound history. A retrieval interpretation similarly treats background cues as access conditions for competing memories. Experimental evidence for renewal without independent contextual inhibition Bouton & King (1983) supports the possibility of modulatory control. The exact terminology varies across associative and cognitive theories, but the common explanatory requirement is that identical target cues can evoke different responses in different circumstances.
An external environment can be represented at several levels. Some models treat context as a set of associative features, others as a hierarchical moderator, and still others as evidence favoring one latent environmental state. These are testable alternatives: a context with strong direct outcome associations should behave differently under specific transfer procedures from a context that merely selects between cue meanings. Generalization gradients, tests with new cues, and independent probes of contextual value can help separate the cases Bouton (2004).
Human experiments make this distinction more complicated because verbal instructions can function as contextual information. An instruction that the rules have changed may induce a rapid representation of a new task state, while a lighting change can shift retrieval without explicit awareness of the manipulation. Vansteenwegen et al. (2005) experimentally demonstrated return of conditioned responses when illumination restored the acquisition context. Such findings motivate multidimensional context definitions that include perceptual, conceptual, and temporal variables, while requiring direct tests before claiming a particular internal representation.
Latent-Cause Inference and Environmental Change
Latent-cause models ask whether prediction errors are attributed to noise within the same situation or to a new underlying state of the environment. Under one state, omission may revise a previously acquired predictive relationship; under inferred state change, acquisition and extinction information can be stored in distinct clusters. A later contextual shift may reactivate the acquisition-related state and produce renewal. This framing provides a computational bridge from contextual modulation to Bayesian inference and can represent why the distribution of omissions, timing, and surprise influence behavioral change Gershman et al. (2013).
The discriminating question is how the organism groups experiences. Repeated cue omissions delivered gradually can produce a different inferred structure from an abrupt transition; reminders and reinstated outcomes can change the posterior probability of earlier causes. Gershman et al. (2013) reported experiments consistent with the proposition that gradual extinction could reduce later fear recovery under their tested conditions. Their study supplies a specified manipulation and formal explanation, but it does not establish that every gradual sequence improves extinction or that the latent causes are directly observed neural entities.
Model comparison requires an independent test of predictions beyond fitting a declining response curve. Investigators can alter temporal spacing, context identity, outcome intensity, or the sequence of reinforcement changes while keeping the amount of experience comparable. They should then evaluate out-of-sample test performance, parameter stability, and whether simpler associative accounts can explain the same data. Latent-cause inference is an explanatory framework for organizing empirical regularities; its value is measured by discriminating predictions, not by redescribing any contextual effect as inference Dunsmoor et al. (2015).
Reconsolidation, Memory Updating, and Erasure Hypotheses
Some extinction procedures are designed to reactivate an established memory before presenting new contradictory information during an assumed reconsolidation window. In animal research, Monfils et al. (2009) reported that a retrieval–extinction arrangement reduced later fear recovery under specified experimental conditions. The human experiment by Schiller et al. (2010) extended this approach with a retrieval reminder followed by extinction, reporting reduced return of conditioned responding. These results raised a mechanistic possibility distinct from ordinary inhibitory retrieval: the original memory might become labile and be modified in a way that reduces future recovery.
The distinction between reconsolidation updating and ordinary extinction is difficult to establish from behavior alone. A reminder trial changes prediction, elapsed time, attention, and the structure of extinction training; any of these might affect retrieval. Boundary conditions include memory age, reminder duration, reinforcement history, species, and precise procedural timing. An apparent reduction in later fear could result from strengthened extinction memory or from different state inference rather than a detectable alteration of the original trace. The theoretical review by Dunsmoor et al. (2015) places the reconsolidation approach within broader questions of competing memory representations.
Replication has been decisive in revising confidence. A preregistered, highly powered, independent direct replication by Chalkia et al. (2020) reproduced critical human procedure conditions but did not find an advantage of reactivation–extinction over conventional extinction for preventing recovery. Its authors also identified reporting problems affecting the original protocol. The evidence therefore supports serious investigation of retrieval-dependent updating while leaving the robustness, generality, and mechanism of this specific human intervention unresolved. A monograph evaluating the literature must give independent failures a place alongside influential initial demonstrations, rather than treating a memorable effect as an established clinical technique.
Instrumental Extinction, Habits, and Avoidance
In instrumental extinction, an action ceases to produce the outcome that previously reinforced it. A fall in action frequency can be governed by response–outcome learning, discriminative contextual control, and response inhibition. The organism's current goal, the cost of responding, and the availability of alternative actions change expression. Todd et al. (2014) reviews convergent evidence that instrumental extinction, like Pavlovian extinction, can be strongly context-dependent and subject to renewal, despite differences in the learning contingencies that initially generated behavior.
Instrumental behavior introduces questions that Pavlovian cue exposure alone cannot answer. If an animal has acquired a habit, omission of reward may suppress the response through new inhibitory control without restoring outcome-value sensitivity. If an avoidance action prevents an aversive event, the absence of that event is compatible with either successful avoidance or genuine safety; learning about safety can require carefully arranged opportunities to experience the outcome independently of the action. Experimental distinctions between expectancy and control of action are therefore essential for predicting behavioral persistence Todd et al. (2014).
Contextual recovery of an extinguished instrumental response has practical analogues in drug seeking, compulsive rituals, and relapse to previously reinforced behavior. These analogues are scientifically useful when the experimental preparation specifies which contingency is reproduced. They do not license a claim that a human clinical disorder is reducible to lever pressing or that every behavioral recurrence shares one associative cause. A robust translation requires outcome-specific measurement, motivational controls, clinical populations where relevant, and longer follow-up than a laboratory test Vervliet et al. (2013).
Amygdala Systems and Fear Expression
Animal research has identified the amygdala as a critical component of acquired defensive responses and their modification. Distinct populations and projections participate in stimulus evaluation, learned threat expression, and suppression of defensive behavior. These findings permit more specific hypotheses than a general claim that the brain stores fear in one region. In a widely cited mouse study Herry et al. (2008), activity in different amygdala circuits was associated with opposing behavioral states following conditioning and extinction. Such circuit-level evidence is informative because it relates identified neuronal populations to behavioral transitions under controlled laboratory histories.
Neural measurements require caution about the meaning of inhibition. Reduced freezing may reflect retrieval of safety information, changes in motor expression, or shifts in arousal, and an inhibitory neuron does not automatically encode a psychological inhibitory association. Conversely, memory competition can be implemented through many distributed pathways. Experiments involving anatomical interventions, selective recording, and timed perturbation provide stronger causal leverage than correlated imaging alone; even then, the exact operation may depend on task stage and training history Dunsmoor et al. (2015).
Extinction therefore involves reorganization of functional control across a network. Evidence of a threat-associated response after later testing suggests that prior learning remained behaviorally accessible, but this inference does not identify whether identical cell ensembles mediated both episodes. Experimental work should compare acquisition, immediate extinction, and delayed retrieval while accounting for the effects of handling, stress, or recording procedures. The functional claim that distinct neural states influence defensive expression is more secure than a universal anatomical mapping of one psychological process to one cell type Herry et al. (2008).
Medial Prefrontal Contributions to Retention
Medial prefrontal cortex has been prominent in extinction research because particular prefrontal regions influence later expression of conditioned fear. In rats, Quirk et al. (2000) reported that lesions of ventromedial prefrontal cortex affected recall of extinction, motivating a distinction between acquisition of response decrement and delayed retention of inhibitory control. Subsequent work examined how patterns of firing during extinction learning predict later retrieval. Such experiments suggest a role for prefrontal regulation of defensive expression while leaving the precise locus of stored information open.
Milad and Quirk Milad & Quirk (2002) showed relationships between neurons in medial prefrontal cortex and memory for fear extinction. Their findings sharpened the hypothesis that prefrontal participation can signal when a learned cue is no longer dangerous under the prevailing circumstances. Nevertheless, distinct subregions and their connectivity must be described according to each species' anatomy and task. A reduced conditioned response following a manipulation cannot establish that prefrontal cortex independently stores a complete safety representation.
During translation to human neuroscience, the commonly used correspondence between rodent infralimbic cortex and human ventromedial prefrontal regions is an anatomical and functional working comparison, not exact identity. Human neuroimaging can detect coordinated activity, while lesion and stimulation approaches provide different causal information and carry their own constraints. Milad et al. (2007) related prefrontal and hippocampal activation to recall of extinction in humans, helping identify a candidate network rather than proving that all anxiety symptoms originate from malfunction of a single region.
Hippocampal Context Processing and Retrieval
The hippocampal formation contributes to representing environmental relations, temporal organization, and context-specific retrieval. Since renewal involves moving between situations that favor different learning histories, hippocampal context processing is a natural candidate for selecting whether acquisition or extinction is expressed. Studies reviewed in Dunsmoor et al. (2015) connect hippocampal and prefrontal contributions to context-sensitive defensive behavior. This account does not require the hippocampus to store every cue–outcome association or to be solely responsible for contextual discrimination.
A rigorous mechanism must specify the context representation's content. Two rooms may be distinguished by visual and olfactory cues; two phases of the same task may differ in time and inferred rule. Hippocampal activity may respond to novelty, episodic structure, spatial differences, or uncertainty as well as extinction retrieval. A causal test must show that manipulation of the implicated circuit changes the relevant context-dependent expression while preserving acquisition and sensory capacities as much as possible Milad et al. (2007).
The behavioral distinction between renewal and reinstatement remains useful at the neural level. Reinstating outcomes may increase threat or reward significance associated with the surrounding environment, while renewal can arise from retrieval competition without demonstrable new contextual excitation. Bouton & King (1983) offers behavioral constraints that a neural explanation must honor. Network findings gain explanatory power when they predict these different functional routes to recovery, rather than grouping every increase in conditioned responding under one generic relapse mechanism.
Human Neuroimaging and Psychophysiological Evidence
Human studies make extinction mechanisms accessible through physiological recording, expectancy ratings, and functional imaging. Phelps et al. (2004) examined extinction learning in the human brain and implicated interactions among amygdala and ventromedial prefrontal processes. Milad et al. (2007) connected extinction-memory recall to activity in prefrontal and hippocampal regions. These experiments provided cross-species bridges but also revealed that an experimental outcome called fear may be operationalized as a combination of skin conductance, subjective expectation, startle, or neural blood-oxygen responses.
The strength of a neural conclusion depends on the measurement design. A blood-oxygen-level-dependent signal is a spatially resolved correlate that integrates complex neuronal and vascular events; it does not identify a signed inhibitory association. Likewise, trial-wise skin conductance is influenced by orienting and nonassociative factors, making cue contrasts and adequate controls important. The methodological audit by Lonsdorf et al. (2017) describes decisions about acquisition criteria, exclusions, psychophysiological scoring, and statistical comparisons that can affect the observed result.
Evidence synthesis is more useful than treating isolated activation coordinates as precise mental functions. Anatomical convergence supports the importance of distributed networks, while variations in stimulus modality, scanner environment, developmental stage, and analysis pipeline reduce one-to-one correspondence between studies. Clinical inference requires an additional step: a network correlated with laboratory fear extinction is not automatically a validated predictor of symptom improvement. Contemporary experimental and review work calls for designs that connect neural retrieval mechanisms with delayed, context-shifted behavioral outcomes Dunsmoor et al. (2015).
Intracranial Recordings and Contemporary Replication
A recent direct human neural contribution came from Pacheco-Estefan et al. (2026), published online in August 2025 and assigned to a 2026 journal issue. The investigators studied forty-nine epilepsy patients with intracranial recordings during a changing-contingency paradigm involving multiple contexts. Their analyses related amygdala theta activity to safety during extinction and identified stable, context-specific representational patterns linked to later memory expression. The study provides uncommon temporal and anatomical access to deep human structures and supports competition between different learned representations as an empirically analyzable mechanism.
These findings require precise qualification. Participants were patients undergoing invasive clinical monitoring, electrode placement depended on clinical necessity, and the laboratory outcome included aversive sensory material and expectancy judgments. The evidence thus supports mechanistic conclusions within the sampled task and implanted regions but does not estimate population prevalence or prove an intervention effect in clinical anxiety treatment. A publisher correction appeared in September 2025, and the updated article should be treated as the bibliographic version of record Pacheco-Estefan et al. (2026).
Replication also tests candidate neuromodulatory mechanisms. A preregistered replication of a dopamine-related extinction-memory hypothesis by Andres et al. (2024) found that the prespecified primary human effect of L-DOPA was not reproduced, while reporting qualified exploratory observations. The scientific implication is methodological: neural plausibility and earlier promising effects do not eliminate the need for independent, preregistered tests. Together, neural recording and replication research support a network-and-retrieval account with substantial remaining uncertainties about causality, generality, and how pharmacological manipulation interacts with learning.
Experimental Paradigms and Control Conditions
A canonical extinction study includes acquisition, extinction, and testing phases, but those labels do not guarantee construct validity. Investigators must define the cue, outcome, intertrial interval, context, reinforcement schedule, response scale, and criteria for demonstrating prior acquisition. An extinction group can be compared with a no-extinction or retention control, while context switches can be crossed with training histories. These factorial choices distinguish general test effects from the recovery of a specific learned response. The published recommendations in Lonsdorf et al. (2017) explain how small design changes can shift what a fear-conditioning experiment actually measures.
Manipulating context requires more than assigning arbitrary background colors. Context A and B should be discriminable, yet sufficiently comparable in arousal, novelty, perceptual salience, and outcome history to support a causal contrast. Context order and cue assignment should be counterbalanced, and baseline responses to contexts should be assessed where feasible. For reinstatement, unsignaled outcome presentations need their own controls; for spontaneous recovery, time-matched test conditions are needed. Studies of human renewal frequently vary these features, which limits simple comparisons across laboratories Wang et al. (2024).
Instrumental experiments require control for alternative reinforcers and response effort. Withholding reinforcement may reduce behavior through satiety, response competition, or changed opportunities rather than a specific acquired inhibitory relation. Yoked procedures, outcome revaluation, context changes, and distinct response keys can identify the controlling contingency. Comparative reviews Todd et al. (2014) show that the conceptual tools of extinction extend beyond fear without making all paradigms interchangeable. A method is informative insofar as its controls eliminate a plausible alternative explanation.
What Physiological and Behavioral Measures Identify
A dependent measure records an observable component of learning or response expression. Freezing in rodents, suppression of appetitive activity, lever pressing, defensive startle, skin conductance, pupil variation, verbal expectancy, and choice each operationalize a different aspect. The same participant may verbally anticipate safety while retaining autonomic differentiation between cues, or physiological activity may decline without equivalent change in reported appraisal. This divergence can reflect separate systems, imperfect reliability, or task-specific demands. Any interpretation of extinction strength must therefore name the measure and specify which construct it is intended to index Lonsdorf et al. (2017).
Skin conductance in particular is sensitive to movement, baseline arousal, electrode placement, habituation, and scoring choices; dichotomizing participants as learners or nonlearners can magnify selection bias. Fear-potentiated startle requires additional probes that may themselves affect learning. Rating scales reveal explicit knowledge but may be influenced by instructions and repeated assessment. Study design should avoid treating agreement among correlated outcome measures as independent replication while seeking convergent evidence from methods with meaningfully different vulnerabilities Craske et al. (2014).
The timing of measurement matters as much as its identity. Within-session slopes measure immediate behavioral adaptation, while delayed probes measure retrieval after time or context changes. Reinstatement tests administered after numerous cue-alone probes can miss recovery that extinguished again during testing. Statistical analysis can address baseline differences, random effects, and participant heterogeneity, but cannot reconstruct an omitted control condition after data collection. The strongest conclusions combine theory-specific contrasts with transparently reported raw trial patterns and reproducible scoring decisions Wang et al. (2024).
Evidence Aggregation and Replication
One influential experiment can reveal a phenomenon; its generality requires evidence across populations, laboratories, and tasks. Reviews must distinguish experiments that demonstrate response return from those that compare methods of preventing it. Wang et al. (2024) combined a systematic review of eighty human renewal studies with meta-analysis of twenty-three studies. It reported the strength of renewal varying across procedural designs and identified contextual manipulations associated with differences in return of fear. That literature is a foundation for designing targeted replications, not evidence that every contextual laboratory arrangement behaves identically.
A second quantitative synthesis Bustamante et al. (2024) evaluated extinction in multiple contexts using twenty-five studies and thirty-seven experimental comparisons. The authors found an overall advantage for multiple-context extinction in reducing later response recovery, with important moderators including conditioned-stimulus exposure duration, species, and recovery paradigm. The evidence includes animal and human preparations and therefore supports a general learning hypothesis more directly than a specific clinical recommendation. The magnitude applicable to particular patients or disorders remains a separate research question.
Registered replication can substantially revise the confidence assigned to proposed procedural improvements. In contrast to the original striking report by Schiller et al. (2010), Chalkia et al. (2020) found no advantage for reactivation followed by extinction in their preregistered direct replication. Different protocols might generate different outcomes, but that possibility should be tested before a broad claim is restored. Transparent sample planning, preregistration, public materials, independent laboratories, and systematic negative-result reporting improve the chance of discovering genuinely robust mechanisms Lonsdorf et al. (2017).
Inhibitory Learning in Exposure-Based Psychotherapy
Exposure-based psychological treatments use repeated engagement with feared cues or situations to alter expectations and response patterns. An inhibitory-learning framework emphasizes opportunities to learn that predicted negative outcomes fail to occur, especially across relevant contexts and time. Craske et al. (2014) described procedural strategies including expectancy violation, variable exposure, retrieval reminders, and reduced reliance on safety signals. These are mechanistically motivated proposals integrated with clinical examples, rather than a demonstration that every strategy independently improves patient outcomes.
A clinically meaningful outcome differs from a laboratory fear index. Symptom severity, functioning, avoidance, distress tolerance, and relapse over follow-up may change for reasons extending beyond Pavlovian inhibitory learning. Psychological treatment also involves interpretation, self-efficacy, therapeutic alliance, goals, and broader social circumstances. The review Vervliet et al. (2013) connected extinction mechanisms to relapse vulnerabilities while documenting substantial limits of translation. Experimental evidence about the return of a conditioned response is therefore an input to treatment research, not a substitute for controlled clinical efficacy trials.
Within clinical research, it is especially important to distinguish mechanism from practice recommendation. An intervention intended to increase expectancy violation should be compared with another credible treatment while measuring the hypothesized mediator, testing maintenance beyond treatment termination, and documenting safety and adverse responses. Population heterogeneity, comorbidity, and feasibility require attention. The monograph's application is explanatory: it identifies why therapists and researchers investigate context, safety behaviors, and inhibitory learning, while clinical decisions properly rest on current guidelines, individual assessment, and qualified care Craske et al. (2014).
Addiction, Appetitive Learning, and Relapse Analogies
Extinction principles apply to appetitive cues and actions as well as aversive conditioning. Environmental signals associated with reward can acquire incentive and predictive functions, while action–outcome contingencies may preserve instrumental responding. When established outcomes are omitted, responding can diminish but later reappear after contextual changes, renewed outcomes, or time. This resemblance to relapse motivates experiments on learned appetitive and drug-associated behavior. The comparative treatment in Todd et al. (2014) shows why instrumental context control deserves consideration alongside Pavlovian cue extinction.
The analogy has clear boundaries. Addiction involves neuroadaptation, motivational shifts, social conditions, access to substances, habit systems, psychiatric factors, and real-world reinforcement patterns that cannot be reconstructed by a simple cue–outcome schedule. A laboratory return of lever pressing or subjective craving is not an epidemiological relapse rate. To examine generalization, investigators need longitudinal outcomes, multiple settings, appropriate patient samples, and designs that separate cue reactivity from actual consumption or functional impairment Vervliet et al. (2013).
An interdisciplinary synthesis can therefore use extinction and renewal as a framework for understanding why environmental change may alter access to earlier learning while recognizing the additional causes of complex behavior. The same logic applies to changes in eating, digital reward habits, or avoidance, but each domain has distinct outcomes and ethical constraints. The scientifically defensible claim is that context-dependent extinction supplies testable hypotheses about behavioral recurrence; its importance must be established by domain-specific experiments and careful clinical validation Bouton (2002).
Optimizing Contextual Transfer and Retrieval
Improving the generality of extinction requires changing the conditions under which a new response pattern can be retrieved. Multiple-context extinction deliberately exposes the learner to the omission contingency across varied surroundings. The rationale is that no single setting becomes the exclusive signal for safety, increasing the opportunity for extinction information to govern behavior elsewhere. The multilevel synthesis by Bustamante et al. (2024) found reduced recovery in the set of experiments using multiple contexts, although effects depended on stimulus exposure, subject type, and recovery test. These moderators constrain how the experimental result should be generalized.
Renewal studies offer complementary evidence. Wang et al. (2024) found that the ABA design was more effective at producing return of fear than certain alternatives and that similarity among acquisition and extinction contexts and training across contexts could weaken renewal. These results suggest that contextual relations should be conceptualized as controllable features of learning and transfer rather than incidental background. However, a procedure that reduces renewal in a brief laboratory task may alter the experience of complexity, novelty, or expectancy in ways that differ from a longitudinal clinical intervention.
Retrieval cues can also reinstate access to extinction information, but their effectiveness depends on what they have learned to signal. A cue that uniquely accompanies nonreinforcement might retrieve the extinction episode, while a cue experienced only in one setting may itself become context bound. Removing excessive reliance on such cues, or varying cue combinations, generates testable predictions. Researchers should jointly assess immediate performance, delayed retrieval, and unfamiliar-context tests to determine whether an intervention strengthened learning, broadened its generalization, or changed an auxiliary response Craske et al. (2014).
Variability, Individual Differences, and Development
The consequences of extinction experience depend on history and current state. Baseline threat learning, reinforcement intermittency, stimulus discriminability, attentional engagement, stress, sleep, and prior exposure can change response trajectories. These variables also interact with analytic decisions because weak initial conditioning leaves little scope for observable reduction. Population contrasts should therefore define acquisition and extinction with comparable measures and report uncertainty around subgroup differences. The methodological standards in Lonsdorf et al. (2017) identify design features that deserve control before a demographic contrast is interpreted as a mechanistic difference.
Across developmental periods, maturational changes in learning, memory, and contextual inference may influence renewal and retrieval, but age-group findings are heterogeneous. The human renewal synthesis Wang et al. (2024) reported reduced sensitivity to contextual changes in some older-adult evidence, while emphasizing overall procedural variability. Such an observation is not a general claim that older people extinguish more effectively; it concerns specific context-change responses under particular task conditions. The broader question is how experience, neurodevelopment, and explicit task understanding interact to influence which competing memory is expressed.
Individual profiles are scientifically meaningful when they predict independent future outcomes. A participant's autonomic response during conditioning, explicit contingency knowledge, and everyday anxious behavior are different variables; a single laboratory index should not be treated as an established clinical biomarker. Samples restricted to healthy university volunteers or invasive-recording patients require special care when generalized to populations with psychiatric symptoms. Mechanistic accounts should specify the level at which variability enters: acquisition, extinction update, context encoding, consolidation, retrieval, or response expression Pacheco-Estefan et al. (2026).
Current Evidence and Outstanding Experimental Disagreements
Recent studies refine the conventional inhibitory-retrieval picture while adding unresolved questions. The intracranial investigation by Pacheco-Estefan et al. (2026) linked context-specific neural representations and amygdala theta activity to the balance between learned threat and safety responses in epilepsy patients. Its result adds direct human neural measurement to a field often reliant on peripheral physiology or hemodynamic correlates. The finding is important because it characterizes representational competition during and after extinction; its sample and context structure also motivate independent tests in broader populations.
A 2026 experiment by Wang et al. (2026) tested variants of human fear extinction involving unpaired unconditioned-stimulus presentations or unconditioned-stimulus-only events. The authors reported limited support for these strategies in reducing contextual renewal, noting that some participants appeared to interpret nominally unpaired presentations as evidence of threat for both cue categories. The study highlights a general principle: experimentally programmed contingencies and participants' subjective models of those contingencies can diverge. A sound theory of inhibitory learning must predict behavior from the information participants infer, not simply the schedule the investigator intended.
Other disagreements remain substantial. Retrieval–extinction interventions have produced both promising original findings and preregistered independent null replications Schiller et al. (2010) Chalkia et al. (2020). Neuromodulatory accounts likewise face negative replication of some prespecified human effects Andres et al. (2024). The review evidence for multiple-context extinction is encouraging Bustamante et al. (2024), but does not resolve optimal dose, long-term persistence, or clinical magnitude. Scientific progress depends on identifying boundary conditions with pre-registered contrasts and meaningful outcome measures, rather than forcing all discrepant results under a single memory-competition label.
Research Program for Mechanistic Discrimination
Future extinction experiments should vary features that competing theories treat differently. A design could compare abrupt and gradual changes in reinforcement while orthogonally manipulating context similarity and the acquisition-to-extinction interval. A latent-cause model would predict differences in inferred state partitioning, whereas a simpler associative model could predict changes based on accumulated prediction error and generalization. Both should submit quantitative predictions before observing test data. The target outcomes would include extinction slopes, immediate retention, renewal under multiple arrangements, and delayed spontaneous recovery Gershman et al. (2013).
Additional studies can test whether extinction information transfers across response systems. For example, expectancy ratings may update while autonomic conditioning remains, or instrumental responding may renew despite changes in explicit outcome value. A preregistered multi-measure design could investigate which relations are shared and which are method-specific. Simultaneous neural recording offers further hypotheses, but causal intervention should be reserved for ethically justified conditions and interpreted in light of species and clinical sampling. Generalizability improves when laboratories share cue schedules, scoring code, exclusion criteria, and trial-level datasets Lonsdorf et al. (2017).
The Artificial Era adds researchable problems of environmental context and cognitive partnership, particularly when digital agents scaffold exposure, supply retrieval reminders, or change the interpretation of learned contingencies. Such systems can be analyzed as parts of the learning environment: their timing, instructions, reliability, and perceived agency may alter prediction and transfer. Whether an artificial system itself develops an analogous extinction representation depends on independently specified architecture and behavioral tests. These are future comparative hypotheses and should be distinguished from the empirical evidence on human and animal conditioning assembled here Todd et al. (2014).
Conclusion
Extinction is a systematic change in the expression of previously acquired learning following alteration or omission of its supporting contingency. The foundational experimental finding is that reduced responding at the end of nonreinforced experience can coexist with spontaneous recovery, renewal after context change, reinstatement following unsignaled outcomes, and rapid reacquisition. These phenomena make extinction a privileged domain for investigating how new information competes with earlier histories. Classical animal experiments, controlled human conditioning studies, and modern reviews converge on the explanatory importance of context and retrieval Bouton (2004) Rescorla (2004a) Wang et al. (2024).
Inhibitory learning provides a productive organizing account, while formal latent-cause models, associative updating theories, and neural circuit approaches specify complementary and sometimes competing explanations. Contemporary human intracranial measurements support context-sensitive representation, yet they do not close the causal or translational questions Pacheco-Estefan et al. (2026). Independent failures to reproduce particular reconsolidation or drug-related effects further show why mechanistic plausibility must be joined to replication Chalkia et al. (2020) Andres et al. (2024). The strongest empirical claims remain attached to clearly defined manipulations, observable outcomes, and explicit comparisons.
The major scientific task is to identify the conditions under which extinction information remains available across time, setting, and action. This requires long-term and cross-context testing, theory-discriminating experiments, transparent psychophysiological methods, independent replication, and cautious translation from controlled conditioning to complex clinical life. Multiple-context training has promising aggregate experimental evidence, but patient-level efficacy and optimal implementation require their own tests Bustamante et al. (2024) Craske et al. (2014). Extinction research thus establishes a general principle of psychological organization: an experience can alter behavioral control while the significance of earlier experience continues to depend on the circumstances of retrieval.
References
- Andres, E., Hu, C.-P., Gerlicher, A. M. V., Meyer, B., Tüscher, O., & Kalisch, R. (2024). Replication study on the role of dopamine-dependent prefrontal reactivations in human extinction memory retrieval. Nature Communications, 15, Article 2699. https://doi.org/10.1038/s41467-024-46936-y
- Bouton, M. E. (2002). Context, ambiguity, and unlearning: Sources of relapse after behavioral extinction. Biological Psychiatry, 52(10), 976–986. https://doi.org/10.1016/S0006-3223(02)01546-9
- Bouton, M. E. (2004). Context and behavioral processes in extinction. Learning & Memory, 11(5), 485–494. https://doi.org/10.1101/lm.78804
- Bouton, M. E., & Bolles, R. C. (1979a). Contextual control of the extinction of conditioned fear. Learning and Motivation, 10(4), 445–466. https://doi.org/10.1016/0023-9690(79)90057-2
- Bouton, M. E., & Bolles, R. C. (1979b). Role of conditioned contextual stimuli in reinstatement of extinguished fear. Journal of Experimental Psychology: Animal Behavior Processes, 5(4), 368–378. https://pubmed.ncbi.nlm.nih.gov/528893/
- Bouton, M. E., & King, D. A. (1983). Contextual control of the extinction of conditioned fear: Tests for the associative value of the context. Journal of Experimental Psychology: Animal Behavior Processes, 9(3), 248–265. https://pubmed.ncbi.nlm.nih.gov/6886630/
- Bustamante, J., Soto, M., Miguez, G., Quezada-Scholz, V. E., Angulo, R., & Laborda, M. A. (2024). Extinction in multiple contexts reduces the return of extinguished responses: A multilevel meta-analysis. Learning & Behavior, 52(3), 209–223. https://doi.org/10.3758/s13420-023-00609-w
- Chalkia, A., Schroyens, N., Leng, L., Vanhasbroeck, N., Zenses, A.-K., Van Oudenhove, L., & Beckers, T. (2020). No persistent attenuation of fear memories in humans: A registered replication of the reactivation-extinction effect. Cortex, 129, 496–509. https://doi.org/10.1016/j.cortex.2020.04.017
- Craske, M. G., Treanor, M., Conway, C. C., Zbozinek, T., & Vervliet, B. (2014). Maximizing exposure therapy: An inhibitory learning approach. Behaviour Research and Therapy, 58, 10–23. https://doi.org/10.1016/j.brat.2014.04.006
- Dunsmoor, J. E., Niv, Y., Daw, N., & Phelps, E. A. (2015). Rethinking extinction. Neuron, 88(1), 47–63. https://doi.org/10.1016/j.neuron.2015.09.028
- Gershman, S. J., Jones, C. E., Norman, K. A., Monfils, M.-H., & Niv, Y. (2013). Gradual extinction prevents the return of fear: Implications for the discovery of state. Frontiers in Behavioral Neuroscience, 7, Article 164. [See published corrigendum, 2021.] https://doi.org/10.3389/fnbeh.2013.00164
- Herry, C., Ciocchi, S., Senn, V., Demmou, L., Müller, C., & Lüthi, A. (2008). Switching on and off fear by distinct neuronal circuits. Nature, 454(7204), 600–606. https://doi.org/10.1038/nature07166
- Lonsdorf, T. B., Menz, M. M., Andreatta, M., Fullana, M. A., Golkar, A., Haaker, J., Heitland, I., Hermann, A., Kuhn, M., Kruse, O., Meir Drexler, S., Meulders, A., Nees, F., Pittig, A., Richter, J., Römer, S., Shiban, Y., Schmitz, A., Straube, B., ... Merz, C. J. (2017). Don't fear 'fear conditioning': Methodological considerations for the design and analysis of studies on human fear acquisition, extinction, and return of fear. Neuroscience & Biobehavioral Reviews, 77, 247–285. https://doi.org/10.1016/j.neubiorev.2017.02.026
- Milad, M. R., & Quirk, G. J. (2002). Neurons in medial prefrontal cortex signal memory for fear extinction. Nature, 420(6911), 70–74. https://doi.org/10.1038/nature01138
- Milad, M. R., Wright, C. I., Orr, S. P., Pitman, R. K., Quirk, G. J., & Rauch, S. L. (2007). Recall of fear extinction in humans activates the ventromedial prefrontal cortex and hippocampus in concert. Biological Psychiatry, 62(5), 446–454. https://doi.org/10.1016/j.biopsych.2006.10.011
- Monfils, M.-H., Cowansage, K. K., Klann, E., & LeDoux, J. E. (2009). Extinction-reconsolidation boundaries: Key to persistent attenuation of fear memories. Science, 324(5929), 951–955. https://doi.org/10.1126/science.1167975
- Pacheco-Estefan, D., Bouyeure, A., Jacob, G., Fellner, M.-C., Lehongre, K., Lambrecq, V., Frazzini, V., Navarro, V., Güntürkün, O., Shen, L., Yang, J., Han, B., Chen, Q., & Axmacher, N. (2026). Representational dynamics during extinction of fear memories in the human brain. Nature Human Behaviour, 10(1), 29–48. [Online 2025; publisher correction issued September 2025.] https://doi.org/10.1038/s41562-025-02268-5
- Phelps, E. A., Delgado, M. R., Nearing, K. I., & LeDoux, J. E. (2004). Extinction learning in humans: Role of the amygdala and vmPFC. Neuron, 43(6), 897–905. https://doi.org/10.1016/j.neuron.2004.08.042
- Quirk, G. J., Russo, G. K., Barron, J. L., & Lebron, K. (2000). The role of ventromedial prefrontal cortex in the recovery of extinguished fear. Journal of Neuroscience, 20(16), 6225–6231. https://doi.org/10.1523/JNEUROSCI.20-16-06225.2000
- Rescorla, R. A. (2004a). Spontaneous recovery. Learning & Memory, 11(5), 501–509. https://doi.org/10.1101/lm.77504
- Rescorla, R. A. (2004b). Spontaneous recovery varies inversely with the training-extinction interval. Learning & Behavior, 32(4), 401–408. https://doi.org/10.3758/BF03196037
- Schiller, D., Monfils, M.-H., Raio, C. M., Johnson, D. C., LeDoux, J. E., & Phelps, E. A. (2010). Preventing the return of fear in humans using reconsolidation update mechanisms. Nature, 463(7277), 49–53. https://doi.org/10.1038/nature08637
- Todd, T. P., Vurbic, D., & Bouton, M. E. (2014). Behavioral and neurobiological mechanisms of extinction in Pavlovian and instrumental learning. Neurobiology of Learning and Memory, 108, 52–64. https://doi.org/10.1016/j.nlm.2013.08.012
- Vansteenwegen, D., Hermans, D., Vervliet, B., Francken, G., Beckers, T., Baeyens, F., & Eelen, P. (2005). Return of fear in a human differential conditioning paradigm caused by a return to the original acquisition context. Behaviour Research and Therapy, 43(3), 323–336. https://doi.org/10.1016/j.brat.2004.01.001
- Vervliet, B., Craske, M. G., & Hermans, D. (2013). Fear extinction and relapse: State of the art. Annual Review of Clinical Psychology, 9, 215–248. https://doi.org/10.1146/annurev-clinpsy-050212-185542
- Wang, Y., Ney, L. J., & Lipp, O. V. (2026). The effect of extinction training with unpaired US and US only presentations on the contextual renewal of human fear conditioning. Psychophysiology, 63(6), Article e70312. https://doi.org/10.1111/psyp.70312
- Wang, Y., Olsson, S., Lipp, O. V., & Ney, L. J. (2024). Renewal in human fear conditioning: A systematic review and meta-analysis. Neuroscience & Biobehavioral Reviews, 159, Article 105606. https://doi.org/10.1016/j.neubiorev.2024.105606