This paper presented two experiments in which the gender composition of groups of college students was varied to examine its impact on performance of women on math and verbal tests. In Experiment 1, women completed difficult items from either the math or verbal portion of the GRE, having been told that their scores would be shared at the end of the testing session. Also in the testing room were either two males or females who posed as fellow test-takers. Women performed more poorly on the math items in the presence of two males compared with two females, but performance on the verbal items was equivalent. In Experiment 2, student participants completed just the math items, but male participants were included and a mixed-sex condition was added (i.e., 1 female and 1 male accomplice). Consistent with Experiment 1, women on average performed more poorly when in the presence of two men, but men were unaffected by the gender composition of the group. Results also showed that women's performance decreased in proportion to the number of fellow male test-takers. These results show that the social context alone, specifically being a numerical minority in a situation where a group-based stereotype might apply to one's behavior, can create stereotype threat and undermine performance.