This study was designed to assess the utility and impact on perinatal mortality of a model traditional birth attendant (TBA) training program in rural Mozambique by comparing birth attendance and outcomes in similar communities with and without trained TBAs. Birth attendants and pregnancy outcomes were compared in 1) communities with good access to trained TBAs, 2) randomly selected, comparable communities with no access to trained TBAs, and 3) communities with good access to functioning maternities. Information was collected by interviews with women in randomly selected households. A total of 4,169 women were interviewed who reported on 3,616 completed pregnancies, which resulted in a birth or fetal death. Among women with good access to trained TBAs, 33% reported giving birth attended by a trained TBA, 43% reported giving birth at a health facility, and 24% reported giving birth attended by an untrained person. Among women without access to trained TBAs, 58% reported giving birth at health facilities, and 42% reported attendance by untrained persons. Among women with access to functioning maternity centers, 77% reported giving birth at a health facility and 22% said their birth was attended by an untrained person. There was no significant difference in perinatal or infant mortality among the groups. This study demonstrated a preference for health facility deliveries among rural Mozambican women with good access to trained TBAs. It also failed to demonstrate a reduction in perinatal or infant mortality associated with TBA training. Women said they preferred to deliver in health facilities because conditions were considered better and interventions could be performed if needed. The preference for health facility birth over home birth with a TBA may have been related to difficulties with TBA neighbors and their families or fear of potential witchcraft. Efforts to promote TBA training should be balanced with support for birthing services based in health facilities.