Transfer of learning describes the influence of knowledge, skills, strategies, or habits acquired in one situation on performance in another. A student demonstrates transfer when a mathematical principle helps solve a differently worded problem, when an understanding of scientific evidence guides interpretation of an unfamiliar claim, or when a professional adapts a practiced procedure to a new setting. The essential research question is whether learning changes what a person can do beyond the conditions in which it was acquired. A correct answer on a familiar exercise establishes performance on that exercise; evidence of transfer requires a consequential change in task, context, representation, or demand. The scientific literature distinguishes the distance between learning and application, the mechanisms connecting them, and the conditions that make generalization more likely. Barnett and Ceci, 2002
Transfer matters because education is organized around future use. Students learn reading, mathematics, scientific reasoning, historical interpretation, and professional judgment partly so that these capacities can inform situations that teachers cannot specify in advance. Yet knowledge is frequently bound to the cues, language, examples, tools, and routines used during instruction. A learner can remember a formula and still fail to recognize when it applies; another can recognize the problem but lack sufficient fluency to execute the method. These are different barriers to transfer and require different educational responses. Experiments on analogical problem solving and comparative instruction show why noticing shared structure is a central challenge. Gick and Holyoak, 198390002-6); Rittle-Johnson and Star, 2007
What Counts as Transfer of Learning?
Transfer is an outcome assessed by comparing what learners can do in a target situation after an earlier learning experience. In a strong experiment, researchers vary the learning experience and compare performance on relevant new tasks against a suitable control or alternative instructional condition. Transfer can be positive, when earlier learning improves subsequent performance, or negative, when an earlier response competes with the appropriate response in the new setting. It can also be absent: successful training may leave performance on the target measure unchanged. This definition demands specificity about the source activity, the target activity, and what counts as improvement. Barnett and Ceci, 2002
Transfer must also be distinguished from retention. Retention concerns whether information or skill remains available after time has elapsed; transfer asks whether it supports performance under changed conditions. A delayed test that repeats essentially the same problem evaluates retention more directly than generalization. A novel problem given immediately after instruction may evaluate transfer but tell us little about long-term durability. The strongest educational evaluations measure both, preferably after meaningful delays and without reproducing the instructional cues. Studies of testing and interleaving illustrate the value of examining novel items rather than only practiced ones. Pan and Rickard, 2018; Firth, Rivers, and Boyle, 2021
Near Transfer and Far Transfer
Near transfer describes improvement when the new task shares many relevant features with the learning activity, such as applying a practiced equation method to new numerical values. Far transfer refers to application across more substantial changes, such as using a strategy trained in a narrow task to improve performance in a different intellectual domain. The distance is relative to the dimensions being considered. A task may be near in procedure but far in social setting, or near in visual format but far in the underlying principle. Barnett and Ceci's influential taxonomy emphasizes that knowledge type, performance demands, temporal separation, physical and social context, and functional setting can all affect the characterization of transfer. Barnett and Ceci, 2002
A simple near–far classification is therefore an initial description rather than an explanation. The same intervention can show transfer to untrained examples and fail when the surface appearance changes, or show a small advantage across representations while failing in everyday use. Researchers should specify exactly which features remain stable and which change. Broad claims about improving general intelligence, creativity, or academic achievement require target measures genuinely independent of the trained routines. This distinction is particularly important in the cognitive-training literature. Sala et al., 2019
Positive, Negative, and Zero Transfer
Positive transfer occurs when an existing representation or procedure helps a person recognize and solve the new problem. Negative transfer occurs when the same representation prompts an inappropriate inference, for example when a previously reliable rule is applied outside its valid conditions. Zero transfer denotes no measurable benefit or cost relative to the relevant comparison condition. Researchers must distinguish an observed null result from proof that transfer is impossible: an insensitive assessment, insufficient sample, weak training, or poorly specified target can obscure an effect. A defensible conclusion always names the source and target tasks, comparator, uncertainty, and scope. Barnett and Ceci, 2002
Why Transfer Often Fails
The most common explanatory family concerns the relation between learned representations and the cues available during later performance. Learning may encode the details of an example while leaving its general organizing principle weakly specified. A learner who solved a task in a classroom with an explicit chapter heading may fail outside class because the new setting does not advertise which method to use. The obstacle is then retrieval and selection as much as missing content. Analogical-transfer experiments show that participants may possess an appropriate prior solution yet fail to use it until they receive a hint or a structural cue. Gick and Holyoak, 198390002-6)
A second obstacle is shallow understanding. Memorizing the final operation can produce correct responses to examples that preserve the same surface structure, while novel problems demand comprehension of the causal, quantitative, or relational model. A third obstacle is insufficient foundation: recognizing a deep analogy still may not help when the required domain knowledge, language, or component skills are unavailable. A fourth obstacle is inappropriate generalization: learners may abstract a rule too widely and use it in cases where its assumptions fail. Instruction for transfer must therefore develop usable representations, recognition of relevant conditions, fluent constituent skills, and the ability to evaluate whether an analogy applies. Rittle-Johnson and Star, 2007; Goldstone and Son, 2005
Surface Similarity Can Hide Deep Differences
Two word problems can feature trains, distances, and speeds while requiring different mathematical relationships. Conversely, a medical-treatment scenario and a traffic-management scenario may share a formal principle even though their familiar nouns differ. Transfer depends on detecting the useful relationship rather than automatically matching the most visible objects. Work on analogical reasoning differentiates surface correspondences from deeper relational alignment and investigates the conditions under which people notice the latter. Gick and Holyoak, 198390002-6)
Surface details are still important. They supply memory cues, connect abstract ideas to experience, and reveal what learners perceive as relevant. Effective teaching helps learners use these details as a bridge toward the underlying structure. In simulations of complex systems, changing the appearance of a learning environment and moving from concrete to idealized forms supported transfer better than maintaining one unchanging representation. This finding supports carefully staged abstraction, while its two-experiment scope counsels against treating a single representation sequence as universally optimal. Goldstone and Son, 2005
Practice Can Become Dependent on Prompts
A problem set that groups identical items together often allows a learner to apply the most recently demonstrated method repeatedly. Such performance may reflect successful execution after the decision has already been made by the lesson structure. In a mixed assessment, the learner must first discriminate which principle or procedure is appropriate. Rohrer and Taylor's experiments found better delayed mathematics performance after shuffled or spaced practice than after conventional blocked or massed practice in the studied tasks; the evidence supports reconsidering practice organization while acknowledging differences between laboratory exercises and full curricula. Rohrer and Taylor, 2007
Theoretical Accounts of Transfer
Identical Elements and Overlapping Representations
One influential account predicts that transfer depends on the overlap between what was learned and what the target task requires. Overlap may occur in stimuli, component procedures, representations, or relations. This helps explain why repeating a practiced skill yields reliable improvement and why apparently distant applications are harder to achieve. Contemporary cognitive-training syntheses are broadly consistent with this constraint: gains on trained and closely related tasks are more common than improvements on distant cognitive or academic outcomes. Sala et al., 2019; Aksayli, Sala, and Gobet, 2019
Overlap alone is not a complete theory. A learner may possess the necessary components and still fail to retrieve or coordinate them. Likewise, transfer can be supported when instruction makes a previously unnoticed structural relation explicit. The scientific problem is to specify which features and operations are causally relevant, rather than equating all visual or semantic resemblance with transfer potential. Gick and Holyoak, 198390002-6)
Abstraction, Schema Formation, and Analogy
A schema is an organized representation of relations, conditions, and actions that can apply across particular examples. Schema formation can support transfer because a new task need not reproduce the superficial details of its training examples. In classic experiments, Gick and Holyoak investigated whether comparing analogs or abstracting a shared structure increased the likelihood of solving a new but analogous problem. Their work established a research program centered on representation and retrieval, rather than a guarantee that every instructed analogy will generalize. Gick and Holyoak, 198390002-6)
Analogy is productive when a learner maps relationships accurately, checks corresponding constraints, and adapts the inferred method to the target. A superficial comparison can also mislead. Teachers can improve the quality of comparison by asking what remains invariant, what changes, which assumptions license the inference, and when the analogy breaks down. In mathematics classrooms, the experimental comparison of alternative equation-solving methods produced gains in procedural knowledge and flexibility in the particular sample studied, while conceptual gains were comparable between conditions. The specificity of that result matters: comparison supported some outcomes more clearly than others. Rittle-Johnson and Star, 2007
Situated Participation and Context
Learning occurs within practices, social expectations, languages, tools, and institutions. These contextual elements influence both the knowledge acquired and the opportunities to display it. A person may be capable of applying an idea in a familiar collaborative environment but perform differently under solitary testing, limited time, unfamiliar terminology, or new cultural expectations. Barnett and Ceci's taxonomy treats context as a multi-dimensional feature of transfer and discourages comparisons between studies that use radically different definitions of distance. Barnett and Ceci, 2002
The educational implication is to connect instruction to authentic application while also deliberately varying environments. Authenticity can supply meaningful constraints and incentives; variation can reduce dependence on one exact setting. How these elements should be balanced depends on the learner's prior knowledge and the goal of the task. A realistic simulation is valuable only to the extent that it trains and assesses the features that matter for later work. Goldstone and Son, 2005
Preparation for Future Learning
Some instruction makes students better at learning from a later opportunity rather than immediately improving an unaided transfer test. In this perspective, the target outcome is the capacity to notice what must be learned, ask productive questions, use resources, or benefit from subsequent explanation. Schwartz and Bransford's work on contrasting cases examined how preparatory activities could make later instruction more meaningful. The resulting research challenges evaluation systems that recognize only immediate independent problem solving as evidence of educational value. Schwartz and Bransford, 1998
Preparation-for-future-learning outcomes still require controlled measurement. An assessment might present students with a novel resource, allow equal time to study it, and test what they can learn from it. Such a design measures a different target from an unsupported application test. These two kinds of transfer should be named separately, because a curriculum can be valuable on one and ineffective on the other. Research comparing invention with contrasting cases against being told procedures first offers one experimental line of evidence, with effects depending on the sequence and the instructional content. Schwartz, Chase, Oppezzo, and Chin, 2011
Abstraction: How Much Concreteness Helps?
Concrete examples are indispensable for making unfamiliar concepts interpretable, yet distinctive surface details can dominate attention and impede recognition of common structure. Abstract representations highlight selected relations while omitting incidental properties, but abstraction can become opaque if learners lack grounding. The relevant design question is how representations help a particular group learn the relevant principle and recognize it elsewhere. Goldstone and Son, 2005
Kaminski, Sloutsky, and Heckler reported an advantage for abstract examples in a specific mathematical learning experiment. That result supports the possibility that highly concrete instruction can impair transfer under certain conditions; it does not imply that mathematics should always be taught without concrete models. Goldstone and Son found benefits from a transition between concrete and idealized simulations. Together, these studies suggest investigating the sequence, number, and structural clarity of representations instead of demanding one universal degree of concreteness. Kaminski, Sloutsky, and Heckler, 2008; Goldstone and Son, 2005
Concrete-to-Abstract Connections
A useful instructional sequence can begin with an interpretable example, identify the variables and relationships responsible for its outcome, represent these features in a more general form, and then return to a different concrete context. The learner should explicitly explain what is preserved and what changes. The approach aligns with experimental findings on progressive idealization, but teachers need to observe whether the abstraction genuinely helps students solve new problems. A diagram that looks elegant to an expert may remain an unexplained symbol to a novice. Goldstone and Son, 2005
Contrasting Cases
Comparing examples that differ in one important respect can expose a principle that students might miss when each case is presented in isolation. Contrasting methods of solving the same algebraic equation can reveal why some operations preserve equivalence, and comparing related scientific scenarios can reveal which variables govern an effect. Experiments on comparing equation-solving methods and studies of preparatory contrasting cases give this principle a concrete empirical foundation. Effective comparison requires students to attend to meaningful relations; placing two worksheets side by side is insufficient by itself. Rittle-Johnson and Star, 2007; Schwartz and Bransford, 1998
What the Experimental Evidence Actually Shows
Transfer research contains evidence from tightly controlled analogical problems, mathematics lessons, scientific simulations, classroom field studies, and quantitative syntheses. Their conclusions are complementary because they test different source-to-target distances. A laboratory experiment may isolate a specific mechanism convincingly while giving limited information about classroom implementation. A classroom trial may be more educationally realistic while introducing variation in teacher behavior, student preparation, attendance, and measurement. A meta-analysis can describe average effects across studies but cannot make heterogeneous definitions of transfer equivalent. Barnett and Ceci, 2002
Analogical Transfer Under Experimental Conditions
The Gick and Holyoak program demonstrated that prior analogs can assist a new problem, especially when experimental conditions support recognition of shared structure. Its enduring contribution is explanatory: possession of an appropriate example and spontaneous application of it are separable events. The target problem must trigger or invite the correct relation. In instruction, explicit prompts to identify the underlying principle may initially help; later transfer assessments should remove those prompts to determine whether the learner independently recognizes the connection. Gick and Holyoak, 198390002-6)
Comparison and Flexible Mathematics Knowledge
Rittle-Johnson and Star randomly assigned 70 seventh-grade students to compare alternative equation-solving methods or reflect on methods individually. The comparison condition showed stronger gains in procedural knowledge and flexibility, with comparable conceptual-knowledge gains. This is valuable causal evidence for a targeted instructional format and population. Its sample size and narrow mathematical domain limit claims about the same intervention producing far transfer in all subjects or ages. Rittle-Johnson and Star, 2007
Interleaving and Novel Exemplars
Interleaving asks students to encounter related categories or problem types in a mixed sequence so that they must discriminate among them. Firth, Rivers, and Boyle's systematic review included 26 studies of interleaving for concept learning and a quantitative subset of 17 studies. It reported benefits both for memory of studied items and for transfer to novel category items, with effects varying by task and strongest advantages in contexts requiring subtle discrimination. Reported effect sizes were not a single universal intervention estimate; the review described values reaching approximately Hedges' g = 0.65 for memory and 0.66 for transfer in the analyzed conditions. Firth, Rivers, and Boyle, 2021
The instructional mechanism is plausible: mixed practice prevents the lesson sequence from silently telling students which category they face. Nevertheless, interleaving may create unnecessary difficulty when students cannot yet perform the constituent procedures or have no basis for distinguishing the categories. Teachers should ensure an initial foundation, then assess identification, selection, and execution on fresh problems after a delay. Interleaving is a tool for particular discrimination and generalization goals, rather than a mandate to randomize every instructional sequence. Rohrer and Taylor, 2007; Firth, Rivers, and Boyle, 2021
Retrieval Practice and Transfer
Retrieval practice requires learners to reconstruct information or solutions without simply rereading the answer. Its transfer benefits have been quantified separately from its well-known effects on remembering practiced content. Pan and Rickard synthesized 192 transfer effect sizes from 122 experiments with 10,382 participants and found a random-effects advantage relative to nontesting reexposure controls, d = 0.40 with a 95% confidence interval from 0.31 to 0.50. The effect varied substantially by transfer type and experimental moderator. Transfer was stronger for some shifts in test format, inference, and application and weaker for other changes in stimulus–response relationships. Pan and Rickard, 2018
The review also evaluated publication bias and found that model intercept estimates were often greatly reduced, sometimes eliminating evidence of positive transfer when favorable moderators were absent. Thus the pooled mean is evidence that some kinds of transfer can benefit from testing; it is not evidence that any recall exercise will produce broad, automatic generalization. Practice questions should elicit reasoning and explanation that correspond to intended future use. The specialized mechanisms and evidence for improving retention through testing belong to Retrieval Practice: Why Testing Yourself Improves Memory. Pan and Rickard, 2018
Cognitive Training and the Limits of Far Transfer
Claims that exercising a narrow cognitive capacity will generally improve intelligence or academic performance deserve particular scrutiny. A second-order meta-analysis by Sala and colleagues integrated evidence across working-memory training and several other training categories, including games, music, chess, and exercise-based games. Improvement occurred on training tasks and some related measures, whereas far-transfer effects were small or null; after adjustments for placebo effects and publication bias, the reported overall far-transfer effect was zero. The article synthesized data from 332 samples involving 21,968 participants. Sala et al., 2019
Aksayli, Sala, and Gobet's Cogmed meta-analysis likewise found near-zero effects for distant academic and cognitive outcomes across 50 studies and 637 effect sizes, while improvement on more similar memory tasks was detectable. These findings challenge broad claims that commercial working-memory exercises automatically improve mathematics, reading, or reasoning. They do not disprove every context-specific learning transfer, because applying a deliberately learned mathematical structure to new mathematical problems is a different claim from global cognitive enhancement. Aksayli, Sala, and Gobet, 2019; Gobet and Sala, 2023
Recent Evidence on Metacognitive Regulation
A 2025 field experiment investigated whether training students to regulate their use of one learning strategy could influence their later regulation of mental effort in a different task. Wirth and colleagues cluster-randomized fifth- and sixth-grade students, originally enrolling 777. Pandemic-related restrictions reduced the analyzed sample to 368. The authors reported small positive far-transfer effects on metacognitive effort regulation relative to an untrained control group. The design is relevant because it explicitly tested movement between strategy domains, while the reduced sample and small effects restrict broad conclusions. Wirth et al., 2025
This finding is compatible with the broader pattern that transfer depends on what is taught and what is tested. Training a decision process that is repeatedly applied across tasks may produce a different outcome from drilling a single perceptual or memory routine. It is also important that previous work in the same program contained mixed replication evidence, as discussed in the 2025 article. The result belongs in the preliminary-to-limited category for broad educational policy, even though it provides valuable experimental evidence for a carefully defined target. Wirth et al., 2025
Strategies for Designing Instruction That Transfers
The strongest instructional approach begins by specifying the future performance that learners should eventually demonstrate. The teacher then identifies the principle, procedure, recognition cues, and prerequisite knowledge needed for that performance. Instruction can integrate explicit explanation, comparative examples, representative variation, supported practice, and independent application. Each element targets a distinct potential barrier. No research synthesis supports treating one of these elements as a universal recipe that guarantees distant transfer. Barnett and Ceci, 2002; Firth, Rivers, and Boyle, 2021
Teach the Conditions for Using a Principle
Students need to learn not only how a method works, but when its use is justified. In mathematics, this includes identifying the structural form of a problem, recognizing relevant quantities, and checking whether the method's assumptions hold. In science, it includes distinguishing the prediction of a model from the observations used to evaluate it. Asking students to explain which feature makes a particular strategy appropriate exposes recognition rules that may remain hidden during repetitive exercises. Such prompts can be faded as independence increases. Rittle-Johnson and Star, 2007
Use Multiple Examples With Meaningful Variation
A set of examples becomes powerful when it shows the same principle instantiated in genuinely different ways. Variation must be chosen deliberately: changing irrelevant surface details can test recognition across contexts, while changing a theoretically relevant condition can reveal the boundaries of the principle. Comparing those examples invites learners to state both the invariant relation and the conditions under which it applies. Evidence from analogical reasoning and scientific simulations supports attention to relational structure and representational variation. Gick and Holyoak, 198390002-6); Goldstone and Son, 2005
Compare Confusable Cases
Learners often overgeneralize a familiar solution to a superficially similar problem. Contrasting near neighbors with different underlying structures helps them see which cues matter. For example, two statistical scenarios may look similar but require different assumptions; placing them together can support discrimination. The most useful comparison asks students to predict, justify, and revise their choices. Evidence for structured comparison comes from experiments in algebra and research on contrasting cases, with effects contingent on what students must notice. Rittle-Johnson and Star, 2007; Schwartz and Bransford, 1998
Combine Guided Learning With Independent Selection
Initial worked guidance can help novices coordinate complex components, but continuing to supply a procedure label with every exercise can conceal whether learners know when to choose that procedure. Instruction should eventually require independent problem identification and solution planning. Teachers can gradually remove method cues, mix problem types, and introduce plausible distractors while preserving opportunities for feedback. Performance on unsupported novel problems then offers a clearer indication of transfer than accuracy under permanent prompting. The contrast between single-method reflection and structured comparison provides one relevant experimental anchor. Rittle-Johnson and Star, 2007; Rohrer and Taylor, 2007
Space Applications Across Time
Spacing makes learners reconstruct previous knowledge after some forgetting and encounter it under varying temporal conditions. It is especially relevant when future application will occur long after initial instruction. Yet delayed success on identical items remains a retention measure unless target conditions change. The best evaluation combines delayed assessment with novel examples or task formats. Detailed treatment of memory consolidation and scheduled review belongs to Spaced Repetition: How the Spacing Effect Supports Long-Term Memory. Rohrer and Taylor, 2007
Make Learners Explain Analogies and Their Limits
Asking why an earlier example is relevant to a new problem can make the common relation explicit. A stronger prompt asks where the analogy stops working. This second question protects against negative transfer by drawing attention to constraints. Learners can compare cases verbally, with diagrams, or through symbolic models; what matters is the accuracy of relational mapping and the ability to justify its use. The classic analogical-transfer experiments and work on representational transitions provide the foundation for this design principle. Gick and Holyoak, 198390002-6); Goldstone and Son, 2005
Give Feedback About Reasoning, Not Only Answers
Feedback can identify whether an incorrect answer arose from missing knowledge, a mistaken selection rule, a faulty analogy, or a simple execution error. Those mechanisms call for different responses. A student who knows how to solve a quadratic equation but selects that procedure for a linear problem needs discrimination practice; a student who selects the right method but makes computational errors needs further component practice. The instructional recommendation follows from the transfer distinction between selection and execution, supported by comparison and interleaving studies. Rittle-Johnson and Star, 2007; Firth, Rivers, and Boyle, 2021
Transfer in Different Educational Domains
Mathematics
Mathematics offers clear examples of both successful and unsuccessful generalization. The same underlying operation can appear in equations, diagrams, narratives, and data tables. Students must identify invariant relations despite changing representation. Experiments on abstract mathematical examples, equation-method comparison, and shuffled problem practice show that representation and sequence affect how students perform beyond the examples on which they trained. But these studies examine different constructs; none alone supplies a comprehensive theory of mathematics education. Kaminski, Sloutsky, and Heckler, 2008; Rittle-Johnson and Star, 2007; Rohrer and Taylor, 2007
An instructional assessment might teach students to compare ratios in one visual format and test whether they can interpret equivalent relations in different tables or practical decisions. Researchers should score identification of the relation separately from accurate computation and examine whether novel task features introduce additional language or domain demands. Such a design can identify which component of apparent transfer failed and avoid attributing every error to deficient generalization.
Scientific Reasoning
Scientific transfer involves applying a model to new observations, systems, or experimental situations while respecting assumptions. A learner might use a principle from one ecosystem to reason about another system with similar feedback dynamics. Simulations can make otherwise invisible interactions available for experimentation, but visual similarity may encourage learners to remember stories rather than system structure. Goldstone and Son's simulation experiments support the instructional value of moving between concrete and idealized representations when the intended outcome is transfer of a shared scientific principle. Goldstone and Son, 2005
Scientific reasoning also requires recognizing when a model is insufficient. Transferring an idea appropriately may mean revising or rejecting it after encountering contradictory evidence. Assessments should reward conditional reasoning and explanation, not merely naming the concept associated with a familiar chapter. Evidence from analogical research provides a basis for including structural explanations and boundary checks. Gick and Holyoak, 198390002-6)
Reading, Writing, and Argument
Skills in reading and writing combine domain-general routines with substantial content knowledge. A student may learn to locate a claim and supporting evidence in one text genre yet struggle when the rhetorical structure, vocabulary, or subject matter changes. Transfer assessment should distinguish recognition of argument structure from comprehension of unfamiliar content. More general evidence from transfer taxonomies warns that changes in knowledge domain, functional context, and performance demands cannot be collapsed into one near–far label. Barnett and Ceci, 2002
Writing instruction can therefore practice the same argumentative operation with varied topics while explicitly identifying which parts of the task rely on domain-specific evidence. A learner who can repeat a writing template without evaluating the strength of evidence has acquired a form but may not have developed portable judgment. Comparisons of structurally similar and dissimilar texts can be useful, provided the teacher evaluates performance on independently written work rather than on copied patterns.
Professional and Adult Learning
The distance from a classroom to the workplace includes new tools, collaborators, timing, incentives, safety requirements, and institutional rules. Simulated training may reproduce some essential demands while omitting others. Consequently, a successful examination cannot establish safe or competent job performance by itself. Transfer assessment should include representative tasks, realistic constraints, and where appropriate observation after training. Barnett and Ceci's contextual taxonomy is particularly useful for specifying the dimensions over which generalization is expected. Barnett and Ceci, 2002
In professional education, learning to recognize when a rule should be withheld can be as important as recalling how to apply it. Case comparison, conditional decision points, and feedback about incorrect generalization can make training more representative of actual judgment. Claims of broad professional competence nevertheless require outcomes collected under genuine practice conditions, rather than satisfaction surveys or immediate memory checks.
Individual Differences, Prior Knowledge, and Context
Transfer depends partly on what learners already know. Advanced learners can recognize abstract relations that novices cannot yet distinguish, while novices may require examples that make relations explicit. Language proficiency, familiarity with the task setting, access to materials, working-memory demands, and confidence with the form of assessment can also influence observed performance. These factors should be measured when possible and considered in design, rather than interpreted as fixed ability to generalize. Context-sensitive conceptions of transfer and controlled work on representations support this approach. Barnett and Ceci, 2002; Goldstone and Son, 2005
Age-group differences cannot be inferred by simply comparing studies conducted with children and adults, because task content, instruction, and measures often change together. The second-order cognitive-training synthesis assessed different age groups and reported that near-transfer effects varied, while far-transfer estimates remained small or absent after its corrections. That result is specific to cognitive-training programs and does not imply that children, adolescents, or adults are incapable of applying academic knowledge across contexts. Sala et al., 2019
Multilingual and Inclusive Assessment
Transfer assessments should measure the intended learning construct. If a mathematics task requires advanced unfamiliar vocabulary, poor performance may reflect language demands as well as mathematical application. If a task assumes experience with a particular cultural institution, unfamiliarity with that context can influence what learners recognize. Accessible instructions, alternative representations when appropriate, and analysis of differential task demands can improve interpretability. The goal is to preserve the target reasoning demand while reducing construct-irrelevant obstacles. This recommendation follows from treating transfer as a change across specific knowledge and context dimensions, not as an undifferentiated personal trait. Barnett and Ceci, 2002
Digital Learning and Generative AI
Digital systems can provide explanations, worked examples, retrieval questions, simulations, and opportunities to compare cases. Their educational value for transfer depends on what learners subsequently do independently. A completed assignment, correct chatbot output, or high engagement score does not establish that the learner can select and apply the underlying principle in a novel situation. An appropriate evaluation would compare students with and without a specified tool, use equivalent assessment conditions, and measure delayed independent performance on both practiced and genuinely new problems. Transfer research supplies this evaluation logic; it does not by itself establish the efficacy of any particular AI tutoring system. Barnett and Ceci, 2002; Pan and Rickard, 2018
Generative AI can change the relationship between task completion and learning. If a system performs analogy selection, abstraction, and explanation for the student, performance metrics may reflect assistance rather than acquired competence. Conversely, a tool that prompts the learner to compare representations, justify a decision, and test an independent solution can support activities implicated by established transfer research. These are hypotheses about instructional design; they require direct controlled tests with learner-level outcomes, error monitoring, and appropriate comparator conditions. Broader multidisciplinary questions about AI and educational development are discussed in Education in the Age of AI: Learning, Motivation, Assessment, and Cognitive Development.
Privacy, equitable access, reliability of generated examples, and protection against misleading explanations matter when adopting digital transfer activities. Student work should not be sent to third-party services without an appropriate legal and institutional basis. Educational institutions need to test the factual correctness of generated examples and examine whether assistance changes students' ability to work without it. These are implementation and governance considerations, and no transfer effect should be attributed to a specific AI tool without direct evidence.
How Researchers Should Measure Transfer
A scientifically interpretable transfer study specifies the instructional source, independent target, distance dimensions, timing, outcome rubric, and comparator before evaluating results. The target should require some shared knowledge while changing features relevant to the intended inference. If the target is identical to the training item, the study mainly measures reproduction. If it requires substantial untaught content, the test may underestimate legitimate transfer. Independent raters, preregistered outcomes, delayed testing, and multiple target formats can strengthen a design. Barnett and Ceci's framework helps researchers describe the target rather than treating far transfer as a single undifferentiated quantity. Barnett and Ceci, 2002
Separate Recognition, Selection, and Execution
A useful assessment can ask learners first which principle applies and why, then require them to execute a solution and explain whether its assumptions are satisfied. Distinct scores help determine whether an intervention improved problem recognition, procedural skill, conceptual explanation, or flexibility. Comparisons of different algebraic methods illustrate why these outcomes need not improve together. Rittle-Johnson and Star, 2007
Include Appropriate Controls and Active Comparators
A control group receiving no instruction may reveal an advantage over doing nothing while failing to distinguish a mechanism from additional time, attention, feedback, or expectancy. Where possible, active comparators should match instructional duration and credible engagement while varying the mechanism under investigation. The cognitive-training literature demonstrates how effect estimates can diminish when publication bias and placebo-related design factors are considered. Sala et al., 2019; Gobet and Sala, 2023
Measure Delayed and Authentic Outcomes
Assessing transfer immediately after explanation may capture recently activated cues. Follow-up after days or weeks asks whether the relevant knowledge remains available, while novel cases examine whether it generalizes. A realistic task can add external validity, but researchers need to document what changed between instruction and application. The interleaving and testing syntheses show that design and target type affect measured benefits, making one-time familiar-item scores insufficient for broad educational conclusions. Firth, Rivers, and Boyle, 2021; Pan and Rickard, 2018
Report Effect Size, Precision, and Heterogeneity
Researchers should report estimates with confidence intervals and describe sample composition, attrition, analytical decisions, and uncertainty. An average effect can conceal practical differences across subjects and tasks. Pan and Rickard reported a pooled transfer effect for retrieval-based testing alongside important moderators and publication-bias analyses. Wirth and colleagues found small effects in a field study whose analyzed sample was reduced substantially by pandemic conditions. These examples show why the size, stability, and generalizability of an observed advantage must all be evaluated. Pan and Rickard, 2018; Wirth et al., 2025
Established Findings, Open Questions, and Contested Claims
The established foundation is that transfer can occur and that its probability depends strongly on the relationship between learning and target tasks. Evidence from analogy, comparison, retrieval, and interleaving supports specific routes to improved application under defined conditions. The evidence also shows that distant generalization cannot be presumed from successful training. Broad cognitive-enhancement claims are particularly weak when tested against independent far-transfer outcomes and rigorous comparators. Gick and Holyoak, 198390002-6); Pan and Rickard, 2018; Sala et al., 2019
Several claims remain conditional. The optimal balance between concrete and abstract examples varies with subject matter and learner preparation. The most useful degree of interleaving varies by concept discriminability and initial mastery. Far transfer of metacognitive regulation has promising field evidence but small observed effects and replication questions. Preparation-for-future-learning outcomes offer an additional measure of educational value while requiring careful experimental operationalization. Goldstone and Son, 2005; Firth, Rivers, and Boyle, 2021; Wirth et al., 2025; Schwartz and Bransford, 1998
A general lesson follows from this evidence: strong transfer claims require strong target definitions. Improvement on another example from the same lesson, improvement on a different test format, adoption of a strategy in another school subject, and improvement in daily functioning describe different achievements. Each may be educationally valuable, but they need distinct evidence. Conflating them exaggerates what a study has demonstrated. Barnett and Ceci, 2002
Practical Framework for Teachers and Curriculum Designers
Begin with an authentic future task and write down the principle or decision that must transfer. Identify prerequisite knowledge and teach the central mechanism through understandable examples. Introduce deliberate variation and comparison so that learners can recognize what is essential. After supported practice, remove labels and cues that give away the appropriate method. Include mixed novel tasks, opportunities to explain and check assumptions, feedback linked to the cause of errors, and later assessments without assistance. This sequence synthesizes evidence from analogical reasoning, representational change, mathematical comparison, interleaving, and retrieval rather than reproducing any single published intervention. Gick and Holyoak, 198390002-6); Goldstone and Son, 2005; Firth, Rivers, and Boyle, 2021; Pan and Rickard, 2018
Teachers can examine three separate questions after a unit. Can students reproduce what was taught? Can they recognize and use it when surface details or representation change? Can they learn productively from a new resource when circumstances are unfamiliar? Together these questions distinguish retention, application, and preparation for future learning. Their answers can guide the next instructional decision more precisely than a single overall test score. Schwartz and Bransford, 1998; Barnett and Ceci, 2002
Frequently Asked Questions
Why do students understand a lesson but fail on the exam?
An exam may change the problem's format, remove cues supplied during instruction, require selection among multiple methods, or demand knowledge that was not practiced together. Apparent understanding during a guided lesson can therefore coexist with weak independent recognition or execution. Teachers should compare mistakes on familiar and novel items, examine the depth of explanation, and determine whether the failure concerns retention, selection, transfer, or new content. Rittle-Johnson and Star, 2007; Barnett and Ceci, 2002
Does practicing many examples guarantee far transfer?
No universal guarantee follows from practice quantity. Variation can improve recognition of shared structure, and comparisons can make principles explicit, but broad transfer still depends on what was represented, what was practiced, and what the new task requires. Quantitative reviews of cognitive training show particularly limited generalization to distant outcomes despite intensive practice. Sala et al., 2019
Is far transfer impossible?
Far transfer is an empirical question defined by the source, target, and dimensions of distance. General cognitive-training programs have produced weak or null far-transfer evidence in rigorous syntheses, yet domain-specific research and a 2025 field experiment report transfer on carefully specified, relatively distant tasks. Both conclusions can be true because the interventions and target outcomes differ. Sala et al., 2019; Wirth et al., 2025
What is the difference between transfer and general intelligence?
Transfer concerns the effect of one learning experience on a specified different performance. Intelligence tests attempt to measure broader patterns of cognitive performance. A training program can improve transfer on a related task without meaningfully changing broad intelligence-test performance. Claims that a narrow exercise increases general intelligence require independent evidence on the broader outcome. Gobet and Sala, 2023
How can teachers test transfer fairly?
Teach an identifiable principle, create fresh target problems that rely on the principle while changing relevant conditions, remove diagnostic cues, measure explanation and execution separately, and repeat assessment after a meaningful interval. Compare results with a suitable instructional alternative where possible. Report how the source and target differ, and examine whether language or contextual demands distorted the construct being measured. Barnett and Ceci, 2002; Pan and Rickard, 2018
Conclusion
Transfer of learning is the bridge between acquired knowledge and its subsequent use. This bridge depends on representations that preserve useful relations, recognition of when those relations apply, sufficient domain knowledge, opportunities to practice discrimination, and assessment that tests performance beyond familiar cues. Research supports meaningful transfer under identifiable conditions while imposing strong limits on claims of effortless generalization. Educational design improves when transfer is made an explicit target, measured separately from immediate performance, and evaluated in the actual settings where knowledge is expected to matter. Barnett and Ceci, 2002; Pan and Rickard, 2018
Related Articles
Cluster Learning and Educational Psychology — the field's broader scientific foundations, cognitive mechanisms, teaching practices, and applications.
Retrieval Practice: Why Testing Yourself Improves Memory — the memory mechanisms and study-design evidence behind retrieval-based learning.
Spaced Repetition: How the Spacing Effect Supports Long-Term Memory — delayed retention and practice scheduling.
Education in the Age of AI: Learning, Motivation, Assessment, and Cognitive Development — wider educational implications of artificial-intelligence technologies.
Worked Examples and Scaffolding: How Guided Instruction Supports Learning — Assess worked examples, scaffolding, fading, and the expertise-reversal effect in teaching new skills.
Prior Knowledge and Misconceptions: How They Shape Learning — Examine how existing knowledge and misconceptions shape understanding, conceptual change, and instruction.
Cognitive Load Theory: Working Memory and Instructional Design — Explore intrinsic and extraneous cognitive load, working-memory limits, instructional design, and research debates.
Multimedia Learning: Principles, Benefits, and Cognitive Limits — Assess combined text, graphics, audio, signaling, segmenting, and cognitive limitations in instruction.
References
Aksayli, N. D., Sala, G., and Gobet, F. (2019). The cognitive and academic benefits of Cogmed: A meta-analysis. Educational Research Review, 27, 229–243. https://doi.org/10.1016/j.edurev.2019.04.003
Barnett, S. M., and Ceci, S. J. (2002). When and where do we apply what we learn? A taxonomy for far transfer. Psychological Bulletin, 128(4), 612–637. https://doi.org/10.1037/0033-2909.128.4.612
Firth, J., Rivers, I., and Boyle, J. (2021). A systematic review of interleaving as a concept learning strategy. Review of Education, 9(2), 642–684. https://doi.org/10.1002/rev3.3266
Gick, M. L., and Holyoak, K. J. (1983). Schema induction and analogical transfer. Cognitive Psychology, 15(1), 1–38. https://doi.org/10.1016/0010-0285(83)90002-690002-6)
Gobet, F., and Sala, G. (2023). Cognitive training: A field in search of a phenomenon. Perspectives on Psychological Science. https://doi.org/10.1177/17456916221091830
Goldstone, R. L., and Son, J. Y. (2005). The transfer of scientific principles using concrete and idealized simulations. The Journal of the Learning Sciences, 14(1), 69–110. https://doi.org/10.1207/s15327809jls1401_4
Kaminski, J. A., Sloutsky, V. M., and Heckler, A. F. (2008). The advantage of abstract examples in learning math. Science, 320(5875), 454–455. https://doi.org/10.1126/science.1154659
Pan, S. C., and Rickard, T. C. (2018). Transfer of test-enhanced learning: Meta-analytic review and synthesis. Psychological Bulletin, 144(7), 710–756. https://doi.org/10.1037/bul0000151
Rittle-Johnson, B., and Star, J. R. (2007). Does comparing solution methods facilitate conceptual and procedural knowledge? An experimental study on learning to solve equations. Journal of Educational Psychology, 99(3), 561–574. https://doi.org/10.1037/0022-0663.99.3.561
Rohrer, D., and Taylor, K. (2007). The shuffling of mathematics problems improves learning. Instructional Science, 35, 481–498. https://doi.org/10.1007/s11251-007-9015-8
Sala, G., Aksayli, N. D., Tatlidil, K. S., Tatsumi, T., Gondo, Y., and Gobet, F. (2019). Near and far transfer in cognitive training: A second-order meta-analysis. Collabra: Psychology, 5(1), 18. https://doi.org/10.1525/collabra.203
Schwartz, D. L., and Bransford, J. D. (1998). A time for telling. Cognition and Instruction, 16(4), 475–522. https://doi.org/10.1207/s1532690xci1604_4
Schwartz, D. L., Chase, C. C., Oppezzo, M. A., and Chin, D. B. (2011). Practicing versus inventing with contrasting cases: The effects of telling first on learning and transfer. Journal of Educational Psychology, 103(4), 759–775. https://doi.org/10.1037/a0025140
Wirth, J., Weber-Reuter, X.-L., Schuster, C., Fleischer, J., Leutner, D., and Stebner, F. (2025). Far transfer of metacognitive regulation: From cognitive learning strategy use to mental effort regulation. Educational Psychology Review, 37, Article 7. https://doi.org/10.1007/s10648-024-09983-x