The activation of negative stereotypes can reduce the performance of group members in a test situation. This phenomenon has been referred to as stereotype threat. Although the effect has been frequently replicated, the cognitive basis remains unclear. Different theoretical explanations are based, for example, on a lower motivation of the threatened group (e.g., because the expectation to perform excellently is compromised) or on an increased working memory load due to stereotype-related thoughts. To disentangle the several possible explanations, data of 300 women from an online study on stereotype threat were entered into a diffusion model analysis. In this study, participants had to estimate whether the arithmetic mean of 16 simultaneously presented numbers was above or below 500. Before the experiment began, half of the participating women were told that the study was about gender differences, while gender was not mentioned for the remaining participants. The results show an increased number of errors in the stereotype threat condition. The diffusion model analysis suggests that this effect is due to liberal threshold settings, hat is, a focus on speed rather than accuracy. Possible relations to a promotion focus orientation are discussed.