Gender gaps in mathematics achievement persist in many contexts and when visible, these gaps are paradoxical. Low-stakes measures of mathematics achievement such as grades and study behaviors favor girls, while gaps tend to reverse on assessments/competitions. We explore whether different impacts of raising performance stakes could be one explanation. Study 1 experimentally manipulated the stakes by imposing a performance-contingent, social-evaluative pressure either: before instruction (n = 66), before testing (n = 61), or none (n = 54). Pressure, particularly when experienced during instruction, reduced learning among girls. In contrast, boys trended toward enhanced learning under pressure. In the absence of pressure, girls exhibited strikingly larger gains in learning. Study 2 drew from a larger dataset (n = 386) to interrogate whether girls' superior learning in the no-pressure context might simply be an artifact of differences in prior knowledge, cognitive resources, or demographic variables, but the effect replicated and was not explained by these factors.