We examine a unique admission system at a flagship Colombian university that tracked admitted students into high- and lower-ability classes in the same majors. In a regression discontinuity design, we find that marginal admits to the high-ability classes were less likely to pass their first-year courses, and less likely to earn a college degree in the long-run. These results suggest that college students may benefit when their academic preparation is high relative to that of their classmates, particularly at schools where there is a substantial risk of failing courses.