We evaluate the effectiveness of small high school reform in the country's largest school district, New York City. Using a rich administrative dataset for multiple cohorts of students and distance between student residence and school to instrument for endogenous school selection, we find substantial heterogeneity in school effects: newly created small schools have positive effects on graduation and some other education outcomes while older small schools do not. Importantly, we show that ignoring this source of treatment effect heterogeneity by assuming a common small school effect yields a misleading zero effect of small school attendance. The following are appended: (1) Regents Examinations; (2) Definition of variables; (3) First stage, likelihood of attending a small high school; (4) Relationship between minimum distance to small schools and average student characteristics, by residence zip code; and (5) Full OLS and IV regression results.