Three experiments addressed the performance of women on mathematics tests under varying contexts. In Experiment 1, women and men who were highly identified with mathematics (i.e., had a history of successful performance and valued performance in math) completed an easy or difficult math exam. Performance was equivalent when the test was relatively easy, but men outperformed women when the test was difficult. In Experiment 2, all participants completed the same set of mathematical items, but they were told either that the test they would (stereotype threat) or would not (control) produce gender differences. Women performed more poorly than men but only in the condition in which gender differences were highlighted. Experiment 3 demonstrated similar effects with a student population with more varied background in mathematics. Thus, it appears that stereotype threat can undermine performance of women in mathematics, particularly when a challenging task is confronted and when stereotypic assumptions of gender differences in performance are not specifically dismissed.