Women underperform on standardized math tests compared to men. However, girls perform as well if not better than boys in math. In this paper, it is proposed that this discrepancy may be explained in part by a process of stereotype stratification, in which targets of a negative in-group stereotype view themselves as a member of a subgroup to which the stereotype does not apply. Two experiments with elementary-school children provide support for this theory. In Study 1, girls placed advanced math pictures with males more often than basic math pictures. In addition, girls rated men as liking and as being better at math than women, but viewed boys and girls as being equal on these variables. In Study 2, girls were more likely to draw a man when told a story about an adult mathematician, but were more likely to draw a girl when told of a child mathematician. The social and educational implications of these findings are discussed.