The gender role hypothesis posits that performance on a cognitive ability test is influenced by whether the test instructions frame the test as measuring a skill that is consistent or inconsistent with the test taker's gender role beliefs. The Bem sex role inventory was used to measure the gender role of female college students, and the group embedded figures test (GEFT) was used to measure their spatial ability. Masculine gender role women scored significantly higher on the GEFT when the test was described as measuring spatial ability, whereas feminine gender role women scored significantly higher when the test was described as measuring empathy. In a second experiment, men did not show the same effects. Implications of individual differences in gender role beliefs are discussed.