Article dans une revue: By analyzing a French program that targeted low-achieving and socially disadvantaged junior high schools we provide evidence that school-based compensatory education policies create sorting effects. We use geocoded original data and a regression discontinuity framework to show that the program decreases the individual probability of attending a treated school and symmetrically increases the probability of attending a private school. The effects are driven by pupils from high socioeconomic backgrounds, resulting in an increase in social segregation across schools.