Health systems in developing countries infrequently implement and evaluate maternal death surveillance. This study identified under-reported and misclassified maternal deaths among women of reproductive age between 1999 and 2004 in a rural service unit in Vellore, India. In-depth interviews, semi-structured interviews and structured questionnaires were used to identify maternal deaths known to health care providers and community leaders who regularly come in contact with pregnant women. Eighteen under-reported and misclassified cases--or 50% of maternal deaths--were reported. These included 29% of abortion-related and 7% of domestic violence-related deaths. Based on this study's fieldwork, the existing death surveillance system detected 100% of the maternal deaths reported by hospital staff; however, it missed most maternal deaths reported by community workers. The latter are more likely than deaths reported by hospital workers to result from abortion and family violence. The existing surveillance system should be augmented with a community-based death surveillance system. This comprehensive approach identified twice as many maternal deaths than previously recorded and could be applied in other settings. Appropriate public health interventions should be initiated to prevent maternal deaths in this community.