We combine administrative data from a regional public university with a novel revealed-preference indicator of student friendships to show that socially connected first-year university students are more likely to be retained into their second year. The impact of friends on retention is statistically and economically significant: each friend raises the probability of retention by about 0.6 percentage points, an effect size roughly equivalent to 66 SAT points. This effect occurs in the presence of a robust set of explanatory variables, including unique indicators of a student's prior commitment to the university, and applies to wide variety of student subgroups.