We use a large-scale survey and field experiment to evaluate a policy that provided information about college- and major-specific earnings and cost outcomes to college applicants in Chile. The intervention was administered by the Chilean government and reached 30% of student loan applicants. We show that the low-income and low-achieving students who apply to low-earning college degree programs overestimate earnings for past graduates by over 100%, while beliefs for high-achieving students are correctly centered. Treatment causes low-income students to reduce their demand for low-return degrees by 4.6%, and increases the likelihood they remain in college for at least four years. To understand the mechanisms driving the effect of disclosure policies we estimate a model of college demand. We find that disclosure changes college choice by reducing uncertainty about earnings outcomes, but that its impact is limited by strong student preferences for non-pecuniary degree attributes.