We use data on elementary-school students to investigate how the home language and other characteristics of a student’s same-grade schoolmates influence that student’s academic achievement. We exploit the availability of multiple cohorts of data within each school to control for endogenous selection by incorporating school fixed effects in the model. We also exploit the longitudinal structure of the data to estimate value-added models of the educational production function. We find that attending an “enclave” school provides a slight net benefit to Chinese home-language students and a large net cost to Punjabi home language students. The results are consistent with a simple peer effects mechanism in which the academic achievement or behavior of peers is much more important than their home language.