This paper uses unique panel data to estimate value-added models of learning production in private and government schools in Andhra Pradesh (India), examine heterogeneity in the value-added of private schools, and identify the sources of learning in these schools. In rural areas there is a substantial positive effect of private schools on English, no effect on mathematics and heterogeneous effects on Telugu for 8–10 year old children; at 15 years, there are significant but modest effects on Telugu, mathematics and receptive vocabulary. In urban areas, there is no evidence of a positive private school effect. Teachers' absence and effort, teaching practices and class size significantly affect learning but teachers' education, tenure and experience do not. Children in private schools report more positive assessments of their school experience. Results correspond closely with comparable experimental estimates.