OBJECTIVE: To conduct the first rigorous evaluation of the long-term effect of the Comprehensive Rural Health Project on childhood mortality in rural Maharashtra. METHODS: Background information and full birth histories were collected by conducting household surveys and interviewing women. Control villages resembling project villages in terms of population size were randomly selected from an area enclosed by two ellipses centred around, but not including, the project area. An equal number of villages and approximately equal numbers of households and women were randomly sampled from both areas. Cox models with robust standard errors were used to compare the hazard of death among children under 5 years of age in project and control villages. FINDINGS: The hazard of death was reduced by 30% (95% confidence interval, CI: 6% to 48%) after the neonatal period in the project villages compared with control villages after adjustment for caste and religion of subjects and for availability of irrigation in the villages. During the neonatal period there was an increase of 3% in the hazard of death, but it was not statistically significant (95% CI: -18% to 29%). CONCLUSION: Our methods provide useful tools for evaluating long-running community-based primary health care programmes. Our findings add to the growing debate on the long-term sustainability of community-based interventions designed to reduce child mortality.