This study focuses on how college students form expectations about various major-specific outcomes. For this purpose, I collect a panel data set of Northwestern University undergraduates that contains their subjective expectations about major-specific outcomes. Although students tend to be overconfident about their future academic performance, they revise their expectations in expected ways. The updating process is found to be consistent with a Bayesian learning model. I show that learning plays a role in the decision to switch majors and that major switchers respond to information from their own major. I also present evidence that learning is general and not entirely major specific.