This paper studies the limits of school choice policies in the presence of residential segregation. Using data from the Boston Public Schools choice system, I show that white prekindergarteners are assigned to higher-achieving schools than minority students, and that cross-race school achievement gaps under choice are no lower than would be generated by a neighborhood assignment rule. To understand why choicebased assignments do not reduce gaps in school achievement, I use data on applicants’ rank-order choices to estimate preferences over schools, and consider a series of counterfactual assignments. I find that half of the gap in school achievement between white and Black or Hispanic students is explained by minorities’ longer travel distance to high-performing schools. Differences in demand parameters explain a smaller fraction of the gap, while algorithm rules have no effect.