A child learns that one barking dog can bite, then hesitates near every large dog in the neighborhood. A pigeon learns that pecking a key of one color produces food, then pecks at nearby colors as well. These reactions look very different, yet both reveal the same adaptive problem: experience is always specific, while life rarely repeats itself exactly.

Stimulus generalization allows learning to travel from a familiar cue to a new one. A response acquired in the presence of one cue appears when a new cue resembles it. Without that transfer, every unfamiliar variation would require learning from the beginning. With too much transfer, however, harmless differences can lose their meaning. Generalization is therefore not simply a failure to notice detail. It is part of the way behavior balances continuity with discrimination.

What Is Stimulus Generalization?

Stimulus generalization is the tendency for a response learned in the presence of one stimulus to occur in the presence of other, similar stimuli. The learned cue is often called the training stimulus; the novel cues used to test transfer are generalization stimuli. As those test cues become less similar to the training cue, responding commonly declines, producing a stimulus generalization gradient.

The idea has several meanings. Generalization can describe the pattern researchers observe, the experimental method used to reveal it, or the process thought to produce it. Mostofsky cautioned that blending these meanings makes the term so broad that it can become a label for almost any transfer of behavior (Mostofsky, 1965).

How It Differs from Related Learning Concepts

These concepts differ in what changes. Stimulus generalization occurs when a learned response appears in the presence of a similar cue. Response generalization occurs when different behaviors serve a similar purpose. Stimulus discrimination is learning that different cues predict different outcomes. Stimulus control describes the broader influence that environmental cues acquire through conditioning or reinforcement.

Generalization should also be distinguished from general conclusions or stereotypes in ordinary language. Psychological stimulus generalization is a technical description of how learned responding varies across cues. Human meaning, language, memory, and social categories can expand the range of relevant similarity, but those complexities should not be collapsed into a laboratory gradient without evidence.

From Pavlovian Conditioning to Stimulus Control

The historical roots of stimulus generalization reach back to Ivan Pavlov’s conditioning experiments. After a response was established to one tone, nearby tones could also evoke it, with responding decreasing as the test tone moved farther from the trained value. Pavlov interpreted this spread in neurological terms. Later learning theorists retained the orderly behavioral pattern while disagreeing about its mechanism.

Within the learning theories associated with Clark Hull and Kenneth Spence, generalization was explained through interacting tendencies to respond and not respond across similar cues. The operant tradition took a more deliberately empirical approach. From that perspective, the question was not what hidden force had spread through the nervous system, but how behavior changed across carefully controlled variations in the environment. The result was a shift from treating generalization only as a theoretical process to studying it as measurable stimulus control (Mostofsky, 1965).

This history still matters. A gradient is evidence that a response varies systematically across stimuli; it is not, by itself, proof of one particular neural or cognitive explanation. Contemporary research can add perceptual, memory, and neural mechanisms, but those explanations must be distinguished from the behavioral pattern they are intended to explain.

How Generalization Gradients Work

A generalization gradient plots response strength against positions along a stimulus dimension. In classic operant studies, pigeons were trained to peck when a particular wavelength illuminated a key. During testing, nearby wavelengths also evoked pecking, but response rates generally fell as wavelength moved away from the trained value. A steep gradient indicates narrow stimulus control; a flatter gradient indicates broader responding across the tested range (Guttman & Kalish, 1956).

Similarity and Discriminability

Similarity cannot be judged from physical measurements alone. Guttman and Kalish found that equal changes in wavelength did not always appear equally different to the animal. The more easily two colors could be distinguished, the less broadly responding transferred between them (Guttman & Kalish, 1956).

The relevant dimension may also be more complicated than researchers first assume. Pigeons trained with an indicator needle generalized across new needle positions, but their behavior reflected both absolute position and relational properties within the displayed range. The study reproduced a familiar gradient using an unconventional cue and showed that stimulus control can be multidimensional even when an experiment appears to vary only one feature (Dobson et al., 1971).

What a Generalization Gradient Measures

A graph can look precise while concealing decisions about measurement. Response rate is common in operant research, but it can combine several components: whether the organism responds at all, how long the first response takes, and how rapidly responding proceeds once it begins. David Thomas and Dorothy Konick separated these measures in pigeons. Correcting response-rate gradients for first-response latency and nonresponse trials changed the gradients only slightly; in their preparation, the main component was the rate of responding after responding had begun (Thomas & Konick, 1966).

That result supports response rate in this particular method, but it also illustrates a broader lesson. A generalization gradient does not exist independently of the response selected, the testing schedule, extinction, reinforcement, or the scale used to arrange stimuli. Researchers must specify what was trained, what changed during testing, and what counted as response strength. Otherwise, the apparent breadth of generalization may partly reflect the instrument used to observe it.

What Shapes the Breadth of Generalization?

A broad or narrow gradient is not a fixed property of a stimulus. Its form emerges from relations among the learner, the training history, the tested cues, and the consequences of error. Physical resemblance matters, but so do the dimensions the learner attends to, the alternatives encountered during training, and the context in which transfer is tested.

Training History and Consequences

Repeated reinforcement of one cue can establish orderly responding around that cue, while differential reinforcement teaches that nearby differences predict different outcomes. Stimulus spacing, the range of training examples, the cues used in testing, and extinction procedures can all change the breadth of responding researchers observe (Guttman & Kalish, 1956; Hanson, 1959; Mostofsky, 1965). A broad gradient may therefore reflect adaptive caution in one environment and insufficient differentiation in another.

No degree of breadth is inherently healthy. Broad transfer is valuable when the cost of missing a relevant cue is high; narrow transfer is valuable when small differences reliably predict different consequences. Effective adaptation requires enough generalization to use prior learning and enough discrimination to remain responsive to meaningful differences.

Psychological, Contextual, and Conceptual Similarity

Roger Shepard argued that generalization becomes more orderly when similarity is measured by how different stimuli seem, rather than only by their physical properties. Across tasks and species, responding often declined in a predictable pattern as psychological distance increased. In this account, the learner is judging whether a new situation is similar enough to an earlier one to carry the same consequences (Shepard, 1987).

Human similarity can also be organized by knowledge. Joseph Dunsmoor and Gregory Murphy describe how fear may transfer through categories, typicality, causal beliefs, and conceptual associations. A dog collar, a park, and the sound of keys may not physically resemble a threatening animal or automobile accident, yet learned meaning can connect them. In category-based research, learning about a typical category member sometimes generalized more broadly than learning about an atypical member, an asymmetry that cannot be explained by physical distance alone (Dunsmoor & Murphy, 2015).

This does not invalidate laboratory gradients. It specifies their boundary. Simple continua reveal basic regularities of transfer; real-world human generalization adds prior knowledge, context, language, and inference. What counts as similar is partly discovered in perception and partly constructed through learning.

Discrimination Training and Peak Shift

Generalization and discrimination are complementary pressures. Generalization carries a learned response toward similar cues; discrimination training teaches that differences matter. In differential training, responding to one stimulus is reinforced (S+), while responding to another stimulus is not reinforced (S-). This experience often steepens or reorganizes the gradient near S-.

Harley Hanson demonstrated a striking effect. After pigeons learned that one wavelength was rewarded and a nearby wavelength was not, their strongest response sometimes shifted beyond the rewarded cue and farther away from the nonrewarded one. This peak shift shows that behavior reflects both what has been rewarded and what has signaled no reward. Hanson’s findings were too complex to explain as a simple pull toward the rewarded cue and away from the nonrewarded cue (Hanson, 1959).

Peak shift is not a universal outcome. Its size and appearance can vary with stimulus spacing, prior discrimination training, and the range of cues used during testing (Hanson, 1959; Mostofsky, 1965). The important principle is broader: learning changes the psychological landscape. The best response may occur not at the exact training value, but at a value that more clearly separates what has been rewarded from what has not.

How Generalization Is Programmed Beyond Training

Basic experiments ask whether responding transfers. Applied work asks how to make useful learning persist across people, settings, examples, and time. Trevor Stokes and Donald Baer argued that practitioners should not merely train a behavior and hope that it generalizes. Generalization can be treated as an outcome that must be assessed and, when necessary, deliberately programmed (Stokes & Baer, 1977).

Their review identified several strategies: train sufficient and varied examples, include stimuli that will also appear outside training, connect behavior with naturally maintaining consequences, vary nonessential features of instruction, and teach mediating or self-management responses that can travel across settings. These are principles for designing transfer, not guarantees that every behavior should spread without limits.

This applied meaning is broader than a laboratory stimulus gradient. It may include transfer across responses, people, environments, or time. The distinction is important: the gradient literature explains how responding changes across cues, while programming research asks how a useful performance can survive beyond the conditions in which it was directly taught.

Fear Generalization: When Protective Learning Spreads Too Far

Fear makes the tradeoff especially visible. Generalizing from one dangerous cue to similar cues can provide protection before there is time for deliberate analysis. Yet when defensive responding spreads to cues that are safe, daily life can contract around avoidance. Reviews of human fear research distinguish perceptual generalization, based on physical resemblance, from broader conceptual or symbolic transfer. Both can help explain why a learned threat response appears in situations that only partially resemble the original event (Dymond et al., 2015; Dunsmoor & Paz, 2015).

Fear Generalization and PTSD

A study of 67 military veterans illustrates the clinical relevance without reducing PTSD to a single mechanism. Rajendra Morey and colleagues compared 32 veterans with PTSD and 35 trauma-exposed veterans without PTSD. They conditioned fear to a moderately fearful face and later tested responses to faces varying in emotional intensity. Participants with PTSD more often misidentified the most fearful face as the conditioned cue and showed biased activity toward higher-intensity faces across several regions involved in perception, salience, and fear learning. The comparison group showed a pattern of brain connectivity more consistent with extending safety learning to the low-intensity cue (Morey et al., 2015).

These findings do not mean that overgeneralization is the cause of every anxiety symptom or that a laboratory gradient diagnoses PTSD. The study involved 67 veterans, compared groups at one point in time, and used a deliberately simplified task. More broadly, a later meta-analysis of 16 studies found heightened conditioned-fear generalization across anxiety-related groups, but the authors framed the paradigm as a research tool for studying processes that may cut across diagnoses, not as a standalone diagnostic marker (Cooper et al., 2022).

Discrimination, Context, and Safety Learning

It is tempting to describe recovery from excessive fear as the elimination of generalization. A more accurate aim is flexible stimulus control. A person must learn not only that some cues predict danger, but also that other cues, contexts, and relationships signal safety. New safety learning may compete with older threat learning rather than erase it (Dunsmoor & Paz, 2015).

This distinction helps explain why change may be context-dependent. A person can respond calmly in a therapist’s office yet feel threatened in a setting that resembles the original event. Learning becomes more flexible when meaningful variations are encountered and differences are made explicit. Clinical applications require individual assessment, however; laboratory findings support principles of learning, not a universal treatment script.

A Few Words from Psychology Fanatic

The barking dog at the beginning of this article leaves the learner with a difficult task. Ignoring resemblance would waste hard-earned knowledge; treating every resemblance as identity would make the world unnecessarily dangerous. Between those errors lies a flexible form of learning: carrying experience forward while continuing to notice what is different.

Stimulus generalization reminds us that behavior is neither tied mechanically to one exact cue nor released indiscriminately into every new situation. It follows a psychological landscape shaped by consequence, attention, context, prior knowledge, and history. Learning becomes adaptive not when it always generalizes, but when it generalizes with enough precision to meet the next moment.

Associated Concepts

• Associative Learning: The broader process through which organisms learn relationships among cues, events, actions, and consequences.
• Classical Conditioning: Explains how a cue acquires predictive meaning and begins to evoke a learned emotional, physiological, or behavioral response.
• Operant Conditioning: Describes how consequences shape behavior and how reinforcement histories influence responding in the presence of particular cues.
• Behavior Modification: Applies behavioral learning principles to change relationships among environmental cues, responses, and consequences.
• Applied Behavior Analysis: Uses systematic observation and behavioral principles to promote meaningful behavior change and encourage learning to transfer across settings.
• Fear Conditioning: Shows how a previously neutral cue can acquire threat value and how fear may spread to similar or meaningfully related cues.
• Avoidance Coping: Explains how short-term relief can reinforce withdrawal when learned fear generalizes across situations.
• Exposure Therapy: Uses structured encounters with feared cues to support new learning, finer discrimination, and more flexible responding across contexts.

References

Cooper, Samuel E.; van Dis, Eva A. M.; Hagenaars, Muriel A.; Krypotos, Angelos-Miltiadis; Nemeroff, Charles B.; Lissek, Shmuel; Engelhard, Iris M.; Dunsmoor, Joseph E. (2022). A meta-analysis of conditioned fear generalization in anxiety-related disorders. Neuropsychopharmacology, 47(9), 1652-1661. DOI: 10.1038/s41386-022-01332-2
(Return to Main Text)

Dobson, Ricardo; Verhave, Thom; Hegge, Frederick W. (1971). Generalization of indicator needle position in the pigeon. The Psychological Record, 21(2), 251-256. DOI: 10.1007/BF03394016
(Return to Main Text)

Dunsmoor, Joseph E.; Murphy, Gregory L. (2015). Categories, concepts, and conditioning: How humans generalize fear. Trends in Cognitive Sciences, 19(2), 73-77. DOI: 10.1016/j.tics.2014.12.003
(Return to Main Text)

Dunsmoor, Joseph E.; Paz, Rony (2015). Fear generalization and anxiety: Behavioral and neural mechanisms. Biological Psychiatry, 78(5), 336-343. DOI: 10.1016/j.biopsych.2015.04.010
(Return to Main Text)

Dymond, Simon; Dunsmoor, Joseph E.; Vervliet, Bram; Roche, Bryan; Hermans, Dirk (2015). Fear generalization in humans: Systematic review and implications for anxiety disorder research. Behavior Therapy, 46(5), 561-582. DOI: 10.1016/j.beth.2014.10.001
(Return to Main Text)

Guttman, Norman; Kalish, Harry I. (1956). Discriminability and stimulus generalization. Journal of Experimental Psychology, 51(1), 79-88. DOI: 10.1037/h0046219
(Return to Main Text)

Hanson, Harley M. (1959). Effects of discrimination training on stimulus generalization. Journal of Experimental Psychology, 58(5), 321-334. DOI: 10.1037/h0042606
(Return to Main Text)

Morey, Rajendra A.; Dunsmoor, Joseph E.; Haswell, Courtney C.; Brown, Vanessa M.; Vora, Avani; Weiner, J.; Stjepanović, Daniel; Wagner, H. Ryan III; VA Mid-Atlantic MIRECC Workgroup; LaBar, Kevin S. (2015). Fear learning circuitry is biased toward generalization of fear associations in posttraumatic stress disorder. Translational Psychiatry, 5(12), e700. DOI: 10.1038/tp.2015.196
(Return to Main Text)

Mostofsky, David I. (Ed.) (1965). Stimulus generalization. Stanford University Press. ISBN: 978-0-8047-0221-8
(Return to Main Text)

Shepard, Roger N. (1987). Toward a universal law of generalization for psychological science. Science, 237(4820), 1317-1323. DOI: 10.1126/science.3629243
(Return to Main Text)

Stokes, Trevor F.; Baer, Donald M. (1977). An implicit technology of generalization. Journal of Applied Behavior Analysis, 10(2), 349-367. DOI: 10.1901/jaba.1977.10-349
(Return to Main Text)

Thomas, David R.; Konick, Dorothy S. (1966). A comparison of different measures of response strength in the study of stimulus generalization. Journal of the Experimental Analysis of Behavior, 9(3), 239-242. DOI: 10.1901/jeab.1966.9-239
(Return to Main Text)