Using data from Duke University undergraduates, we make three main contributions to the literature. First, we show that data on earnings beliefs and probabilities of choosing particular occupations are highly informative of future earnings and occupations. Second, we show how beliefs data can be used to recover ex ante treatment effects and their relationship with individual choices. We find large differences in expected earnings across occupations and provide evidence of sorting on expected gains. Finally, nonpecuniary factors play an important role, with a sizable share of individuals willing to give up substantial amounts of earnings by not choosing their highest-paying occupation.