Prototype vs exemplar theory: one average or many remembered examples

Prototype theory and exemplar theory are two explanations of how you put a new thing into a category. In prototype theory, you compare the new thing with one summary of the category, the prototype. In exemplar theory, you compare it with many specific examples that you remember, the exemplars. Prototype theory grew from Eleanor Rosch's research in the 1970s. Exemplar theory is usually traced to the context model of Douglas Medin and Marguerite Schaffer (1978).

Most summaries give the two definitions and a bird example. This article also gives you one question that sorts any example, the experiments with their real numbers, and the current state of the debate.

• Prototype theory: the category is stored as one summary, an average or a best example. The prototype does not have to be a real member of the category.
• Exemplar theory: the category is stored as a set of remembered members. Nothing is averaged.
• The sorting question is: is the new item compared with one summary, or with many stored examples?
• Both theories explain why typical members are easy to classify. Researchers needed special experiments to tell them apart.
• Neither theory won. The research shows that the process that dominates depends on the category and on the stage of learning.

The difference in one table

The two theories agree that categories have typical and less typical members. They disagree about what your memory keeps.

Prototype theory: one summary per category

The APA Dictionary of Psychology defines a prototype as "the best or average exemplar of a category". Its example is the prototypical bird: "some kind of mental average" of all the birds that a person knows. The prototype model says that people form this average and then use it to decide whether a new item belongs to the category.

The prototype is not always a real bird. It is a summary that keeps the features that most birds have, such as wings, feathers and flight. A robin has most of these features, so a robin is close to the prototype. A penguin has fewer of them, so it is far from the prototype, but it is still a bird.

Family resemblance: Rosch and Mervis (1975)

The older view of concepts said that every member of a category has the same defining features. Eleanor Rosch and Carolyn Mervis tested a different idea. Philosopher Ludwig Wittgenstein argued in 1953 that members of a category can share a family resemblance. In a family resemblance, each member shares some features with some other members, but no feature is common to all members.

In their first experiment, Rosch and Mervis (1975) used six common categories: furniture, vehicle, fruit, weapon, vegetable and clothing. They took 20 members of each category, from the most typical to the least typical, and 400 psychology students listed the attributes of these items. The researchers then measured how many of an item's attributes the other members of the category shared.

The results were strong:

• The correlation between family resemblance and typicality was between 0.84 (vegetable) and 0.94 (weapon) for the six categories.
• The five most typical furniture items had 13 attributes in common. The five least typical items had 2.
• In the furniture list, chair was the most typical item and telephone was the least typical. In the fruit list, tomato was 19th of 20.

Rosch and Mervis described their study as an empirical confirmation of Wittgenstein's argument. Items do not need a shared definition to form a category. The most typical items are the items that share the most features with other members of the category and the fewest features with other categories. The APA Dictionary entry for family resemblance says that the prototype is based on these members.

The typicality effect

A typical member is easier to classify. The APA Dictionary calls this the typicality effect: people judge more quickly that a dog is a mammal than that a whale is a mammal. Reaction-time studies by Edward Smith, Edward Shoben and Lance Rips in the early 1970s found the same pattern. People needed less time to verify that a typical member, such as a robin, belongs to its category.

Prototype theory explains this result directly. A robin is close to the bird prototype, so the match is fast. A whale is far from the mammal prototype, so the match takes longer.

A prototype that nobody saw

Classic evidence came from studies with dot patterns and other artificial shapes. Michael Posner and Steven Keele did some of the first studies in 1968. In these studies, participants learned categories from distorted versions of a prototype. The researchers never showed them the prototype itself.

Medin and Schaffer (1978) summarise two main results of this research:

1. 1.Participants classified the prototype that they had never seen "virtually as fast and accurately" as the patterns that they had studied.
2. 2.After delays of several days, participants forgot the studied patterns more than the prototype.

The obvious explanation is that the participants made an average of the patterns during learning. Medin and Schaffer wrote that these results "seem to provide cogent evidence for prototype formation". Then they showed that a theory with no stored average could explain them too.

Exemplar theory: every remembered example counts

The APA Dictionary defines exemplar theory as the hypothesis that "categorization depends on specific remembered instances of the category". In this theory, there is no abstract prototype and no rule. The dictionary also gives a second name for it: instance theory.

In everyday English, an exemplar is an excellent example. In this theory, the word has a wider meaning: an exemplar is any member of the category that you remember, typical or not.

Medin and Schaffer's context model (1978)

Medin and Schaffer called their exemplar theory the context theory of classification. In their words, "a probe item acts as a retrieval cue". A new item makes you retrieve the stored items that are similar to it, and the category with the most similar stored items wins.

The model has one important detail. Similarity is multiplied across features, not added. Thus a stored item that differs from the new item on one feature has a strong effect. A stored item that differs on two or three features has a much weaker effect. Because of this rule, the number of very similar stored items is the most important factor.

This rule also explains the prototype result. Medin and Schaffer wrote that the prototype "almost always is the pattern having the greatest number of highly similar category exemplars". The prototype is easy to classify because it resembles many stored members. The mind does not need to store the prototype.

Medin and Schaffer then designed an experiment in which the two types of theory predict different results. In Experiment 2, 32 volunteers learned two categories of geometric shapes. Each shape had four features with two values each, written as 1 or 0. The prototype of Category A was 1111.

• Stimulus 4 (1110) had three features of the prototype. Prototype models predicted that it was the easier item.
• Stimulus 7 (1010) had only two features of the prototype. But it was very similar to two other Category A items, and to no Category B item. The context model predicted that it was the easier item.

The result supported the context model. During learning, participants made an average of 4.9 errors on Stimulus 4 and 3.3 errors on Stimulus 7. The difference was statistically significant. In the later test, participants never correctly identified the prototype (1111) as a new item, although they had never seen it. Over four experiments, the authors concluded, the context theory "consistently gave a better account of the data".

Nosofsky's generalized context model (1986)

Robert Nosofsky extended the context model in 1986. His generalized context model also assumes that people store category exemplars in memory and classify a new item by its similarity to them. It adds selective attention: people give more weight to the features that are important for the category. Thus the same two items can seem more similar or less similar, as a function of the task.

One question that sorts any example

Textbook examples often use words such as "typical" or "example" for both theories. Ask one question: is the new item compared with one summary, or with many stored examples?

The last row comes from the APA Dictionary. It says that exemplar theories of prejudice see stereotypes as memories of "specific individual group members", not only as abstractions. The representativeness heuristic is related: it judges probability by resemblance to a typical case. The two theories ask how that typical case is stored.

Which theory is right?

Neither theory won. The research found that each process works better in some conditions.

Smith and Minda (1998) tested both models at different stages of learning. With smaller categories that were less clearly separated, "the exemplar model dominated even early in learning". With larger categories that were more clearly separated, the prototype model "had a strong early advantage that gave way slowly". The authors suggest a transition from a prototype process to an exemplar process during learning.

A 2014 review by J. David Smith describes the same pattern across many studies. Different processes dominate at early and late stages of learning, with small or large categories, and with well or poorly structured categories. Some researchers therefore accept multiple categorization systems. Others still prefer one exemplar-based process for all categorization.

Smith also compared the two processes in computer simulations. The results measure how well each process performs, not how well each model fits human data:

• In categories with a family-resemblance structure, the prototype process performed better by about 4%. There was never an exemplar advantage.
• In an exclusive-or structure, each category had two separate subgroups, so the two prototypes were almost at the same position. The prototype process stayed at chance (50% correct). The exemplar process reached 86% correct.

Typicality alone does not decide the question either. James Hampton's 2006 review notes a study from 1983 in which people also rated the typicality of members of the category "odd number". Odd numbers have an exact definition. Hampton concludes that typicality differences alone may be "neither here nor there": prototype concepts show them, but other types of concept can show them too.

What AP Psychology asks for

If you study for the AP Psychology exam, learn the College Board's wording. The course and exam description (effective fall 2025) says in Topic 2.2: "Concepts form the basis of thought. Prototypes are the ideal example for any given concept." The same topic covers schemas, assimilation and accommodation, algorithms and heuristics. The description does not mention exemplar theory.

Avoid three common mistakes:

1. 1.A prototype in psychology is not a first model of a product. It is the centre of a category in your mind.
2. 2.A prototype does not have to be a real member. The APA Dictionary calls it a "mental average". Hampton also argues that an abstract summary is a better description than one best example.
3. 3.An exemplar in this theory is not a perfect example. It is any member of the category that you remember.

How to keep the two terms apart

Use a short memory aid. A prototype is one picture of the whole category. Exemplars are the many examples that you remember. Then connect each term to one experiment: Rosch and Mervis's furniture lists for prototypes, and Medin and Schaffer's Stimulus 4 and Stimulus 7 for exemplars.

Concepts are part of semantic memory, your general knowledge of facts, words and concepts. A good test of a psychology term is whether you can recognise a new example of it. Test yourself with new examples instead of reading the definitions again. Our guide to memorising psychology terminology explains why tests help more than a second reading.

Psychology World is a dictionary app with more than 18,000 psychology terms, and it supports this type of practice. Search the dictionary for "prototype", "exemplar theory", "typicality effect" and "family resemblance", and save the terms to one bookmark list. Then study the list as flashcards, as multiple-choice questions, or type each answer. The multiple-choice mode needs a list with a minimum of four terms. The home page shows the four steps of the learning loop and what each study mode does.