We show that the existence of gender differences in performance is highly sensitive to the task used to measure performance, to existing stereotypes and to informational conditions. Out of sixteen purposely designed treatments we find that women underperform when competing only when two conditions are met: 1) the task used is perceived as favoring men and 2) the presence of a rival is strongly primed through information provided before competing. Such sensitivity sheds light on the contradictory evidence found on stereotype-threat causing gender differences in performance under competition.