Three experiments investigated whether men and women have different concerns when confronting a difficult math test that might account for stereotype threat effects. In Experiment 1, male and female undergraduates completed a math test after being told either that the test was designed to "help us identify people who are exceptionally strong in their mathematical reasoning abilities" (hypothesized stereotype threat for men), or "to identify people who are especially weak in their mathematical reasoning abilities" (stereotype threat for women). Women in the latter condition performed more poorly than women in the former condition, but the opposite was true for men. These results suggest that both men and women were affected by stereotype threat when they believed that the gender stereotype applicable to their group was being assessed. Experiment 2 framed the test as being used to identify individuals weak in math (stereotype threat for women) but also introduced for half the participants a possible explanation for failure (i.e., a computer crash that precluded practice before taking a math test). Men were not affected by the computer failure, but women performed better if they had been precluded from practicing on the computer than if they had not been given this attribution for difficulty. Experiment 3 showed that a frame suggesting the test was being used to identify individuals exceptionally strong in math (stereotype threat for men) reversed the effect. Men performed better if they had been provided with an external attribution for failure, but women were not affected by this manipulation. These results suggest that men and women might face different specific types of stereotype threat. For men, the concern might be on meeting the positive stereotype that their gendered group is expected to do well in math. Although one might expect benefits when a positive stereotype is endorsed, pressure for the individual to meet a group-based stereotype might undermine performance. For women, their concern might focus on avoiding the stereotype that women perform poorly in math. In either case, providing an external attribution for failure appears to attenuate stereotype threat effects and, indeed, to improve performance.