Psychologists note that humans use categories to simplify and speed the process of person perception. Applications that require a robot to interact with a variety of different people will demand the creation of stereotypes representing different categories of people. This article presents a method for stereotype learning and usage by a robot. Both robot and simulation experiments are used to examine the benefits and challenges associated with stereotype usage. Our results indicate that stereotypes can serve as a means of feature selection and for inferring a partner’s appearance from observations of their actions. The results also show that the timing of certain types of errors impact the stereotype creation.