When subjects are asked to simply list members of a category (i.e. "make a list of all the birds that come to mind"), they tend to list prototypical members first. That is, they will list "robin" before they list "penguin," suggesting that perhaps the robin is more closely associated with the category "bird" than the penguin is. Also, when subjects are asked to rate how "good" a category member each item is, they tend to rate some birds as being more "birdish" than others, suggesting that they can imagine a typical bird, not just a number of different winged creatures that happen to fit into a category together. Other evidence comes from reaction-time studies of sentence verification. Subjects are shown a sentence and are asked to respond true or false as quickly as possible. Their response time indicates how easy it was for them to decide. Subjects respond more quickly to the sentence "robin is bird" than "penguin is bird." Thus, it was easier for subjects to access the category "bird" and the exemplar "robin" and to decide that they did indeed belong together, suggesting again that "robin" is more closely associated with the category "bird" than is the more peripheral exemplar "penguin." Thus, it seems that the robin may be a prototypical bird, while the penguin may not.

Prototypes require a good deal of experience with the category. If a person has not encountered many exemplars in a certain category, he cannot decide what is typical or atypical of that category. Only after seeing many different members of the category can he have some sense of what is normal for its members. The prototype model also does not establish clear boundaries for the categories. It defines the center, not the bounds, and so allows indistinct boundaries and graded membership. This fits somewhat with available evidence, which suggests that category membership can be influenced by context.

However, it is unclear how the identification system would know which features were important for similarity. Is it important that the new object have four legs? Is it important that it be wooden? There is no simple way, in this model, to distinguish important characteristics from superficial ones.

Interestingly, autistics seem to follow the template model in storing and classifying objects. Alex Michaels, a highly intelligent woman with autism, believes that she, and other people with autism, have no natural method of forming categories. Instead of having a central category called "cat," with many different members, she has separate templates for "Mom's cat" and "Billy's cat." In fact, she is even unable to recognize the same cat in different places, since a template may specify place, as in "the orange cat at Molly's house." As a result of their unusual system of individual classification, autistics have trouble generalizing, forming concepts that include many different examples. Instead of automatically knowing how to form an efficient categorical system, as most people do, autistics must explicitly learn categories; for Alex, that meant studying many pictures of dogs and trying to figure out what important features they had in common. Categorizing via resemblance in this manner is more difficult than it might seem--what is similar about a chihuahua and a Saint Bernard?

Wittgenstein, a twentieth-century philosopher, pointed out the difficulty of creating a reliable definition for any concept, even one that we use easily and often, such as the concept of a "dog." What defines a dog? We might say that dogs (a) are mammals, (b) have four legs, (c) bark to communicate, and so forth. However, there are always exceptions to such definitions that we still consider members of the category. Consider a dog that is born with only three legs, or that has lost a leg in an accident; this animal does not meet requirement (b) of our definition, but we would still consider it a dog nonetheless. Or, consider the Egyptian Basenji, an exotic breed of dog that does not bark. Despite its failure to fulfill criterion (c), we still call it a dog. There seem to be no necessary and sufficient characteristics in our mental definitions. Rather, our concepts appear to follow a rule of "family resemblance"-–all category members resemble each other to some extent, even though each may be missing a different dog-like trait. Any two dogs that we consider will have a great deal in common. However, exactly what they have in common will differ from dog to dog. The rule of family resemblance makes our categories more flexible and more able to deal with the extraordinary variety of animals and objects in the world.