The first study investigated how the purpose for which information about a person is to be used affects the way a perceiver organizes the information. Subjects were asked to categorize and label episodes which described the behavior of a fictional person “Jill” in 64 different situations, and to summarize what each category meant to them. Half of the episodes were easily categorized according to the traits that Jill manifested, half according to features of the situations. One group (personality impression) expected later to describe Jill's personality, a second group (prediction) expected to make predictions about Jill's behavior, and a third group (recall) expected to be tested on their recall of the episodes. The results indicated that subjects in the behavior prediction group categorized the episodes primarily in terms of the personality characteristics Jill portrayed, just as the personality impression group did. In contrast, only the categories formed by the recall group paralleled the built-in structure of the episodes, i.e., they were as often based on features of the situation as on Jill's personality characteristics. A second study showed that the categorization strategies of the recall subjects actually did produce higher recall than those of the personality impression and behavior prediction subjects. The results were interpreted as suggesting that when making behavioral predictions, just as when forming impressions, the layman, like the traditional trait psychologist, prefers to organize information in terms of personality constructs rather than in a way that facilitates retrieval of everything that happened in specific situations. The cognitive costs as well as the gains produced by this strategy merit further scrutiny.