Understanding how returns to higher education vary across degree programs is critical for effective higher education policy. Yet there is little evidence as to whether all degrees improve labor market outcomes, and whether they do so for students from different types of backgrounds. We combine administrative and archival data from Chile with score-based admissions rules at more than 1,100 degree programs to study how the long-run earnings effects of college admission depend on selectivity, field of study, and student characteristics. Our data link admissions outcomes for 30 cohorts of college applicants to administrative records of labor market outcomes up to 30 years post-application. We estimate regression discontinuity specifications for each degree, and describe how threshold-crossing effects vary by degree type. In addition, we use variation in admissions outcomes driven by threshold-crossing to estimate a simple model that maps our discontinuity estimates into causal effects of admission by degree. Observed choice and survey data indicate that the assumptions underlying this model are consistent with student behavior. We find that returns are heterogeneous, with large, positive returns to highly selective degrees and degrees in health, science, and social science fields. Returns to selectivity do not vary by student socioeconomic status. Our findings suggest a role for policies that guide students toward higher-return degrees, such as targeted loans and better college preparation for students from low-income backgrounds.