Using a randomized, between-subjects experimental design, we tested hypotheses that self-critical perfectionism would moderate the effects of subtle stereotype threat (ST) for women and students in underrepresented racial/ethnic groups who are pursuing traditional degrees in science, technology, engineering, or math (STEM). A diverse sample of freshmen students (N = 294) was recruited from 2 major universities. Students were blocked by gender and race/ethnicity and randomly assigned to experience subtle ST or no ST. Participants in the subtle ST condition were primed to consider their gender, race, and ethnicity prior to completing measures of science self-efficacy. Those in the control condition completed the measures without such priming. Controlling for prior academic performance and university context, ST priming significantly interacted (a) with self-critical perfectionism to predict coping self-efficacy scores and (b) with race/ethnicity to predict end-of-semester STEM grades. A 3-way interaction of ST priming, sex, and self-critical perfectionism also predicted students' grades in courses wherein women and men were more proportionally represented. The Sex x Self-Critical Perfectionism interaction was not significant for those in the ST group but was for those in the control group. Men in the control group had higher grade-point averages (GPAs) at low levels of self-critical perfectionism than they had at higher levels of perfectionism. In contrast, women had lower GPAs when self-critical perfectionism was low, but their GPAs were higher when self-critical perfectionism was high. The findings are discussed in terms of their implications for self-efficacy and performance in the pursuit of a STEM major. (Contains 2 tables.)