Two experiments investigated 3- to 5-year-olds' inductive generalizations about social categories. In Experiment 1, participants were shown pictures of children contrasting in appearance and either gender or classmate status, and were asked to generalize either biological properties or behaviors. Contrary to expectations, performance did not differ for chance for gender, but children generalized on the basis of appearance more than classmate status. Experiment 2 further examined children's use of gender for inductive inferences. Children were asked to generalize either stereotyped behaviors (stereotype condition) or novel behaviors (neutral condition) and novel biological properties (both conditions). In the stereotype condition, children generalized both behaviors and biological properties on the basis of gender more than appearance, but, in the neutral condition, children's performance usually did not differ from chance. The implications of these results for essentialism, similarity-based induction, and cognitive variability are discussed.