Standard observed characteristics explain only part of the differences between men and women in education choices and labor market trajectories. Using an experiment to derive students' levels of overconfidence, and preferences for competitiveness and risk, this paper investigates whether these behavioral biases and preferences explain gender differences in college major choices and expected future earnings. In a sample of high-ability undergraduates, we find that competitiveness and overconfidence, but not risk aversion, are systematically related with expectations about future earnings: Individuals who are overconfident and overly competitive have significantly higher earnings expectations. Moreover, gender differences in overconfidence and competitiveness explain about 18 percent of the gender gap in earnings expectations. These experimental measures explain as much of the gender gap in earnings expectations as a rich set of control variables, including test scores and family background, and they are poorly proxied by these same control variables, underscoring that they represent independent variation. While expected earnings are related to college major choices, the experimental measures are not related with college major choice.