This paper studies experimentally anticipated discrimination across gender, hiring patterns, and performance in tasks with different stereotypes in a labor-market setting. Participants are assigned to a seven-people group and randomly allocated a role as a firm or worker. In each group, there are five workers and two firms. The only information firms have about each worker is a self-selected avatar (male, female or neutral) representing a worker's gender. Each firm then decides which worker to hire. Female workers anticipate discrimination when they know the task is math-related, but not otherwise. Men choose similar avatar patterns regardless of the task. Surprisingly, we find no evidence whatsoever of discrimination against females in hiring; in fact female avatars are more likely to be hired. Men do perform at much higher levels in the math-related task, but there is no difference in performance in the emotion-recognition task, where there is a strong female stereotype.