The results from a prospective community survey, a sisterhood method survey, and a hospital survey were compared in order to ascertain a reliable and inexpensive method for estimating direct deaths from obstetric complications of pregnancy. The maternal mortality ratio was used to express risk of dying during pregnancy. The surveys were conducted in Kwimba District in Mwanza region of northwestern Tanzania: in August 1989 to March 1991 in the community study within the primary health care area of Sumve Hospital, which supplied data on maternal mortality between 1986 and 1990. The sisterhood survey was conducted in 2 villages in 1990, of which 1 village was included in the community survey. The village study included 447 women, of whom 421 remained in the survey and delivered 427 infants (415 live born); there was 1 maternal death. The sisterhood method engaged 2865 respondents and the lifetime risk of maternal death was estimated at 297 and the proportional maternal mortality rate was 13.9%. There were 82 maternal deaths and 589 deaths from all causes among sisters aged 15 years and older. 7526 women were included in the hospital survey, of which 7335 births were represented; there were 62 maternal deaths. The maternal mortality risk was 845 among hospital admissions. 69% of all maternal deaths were accounted for by direct causes. Most deaths were attributed to the top 5 worldwide causes: obstructed labor, puerperal sepsis, postpartum hemorrhage, complications of abortion, and preeclampsia. There were few reports of abortions and abortion-related mortality. Relapsing fever or Borrelia infection was an indirect cause of death common to the region and particularly hazardous to pregnant women. Many hospital deaths were emergency admissions. The conclusion was that the sisterhood method provided a better indication of the extent of maternal mortality within the community. Other advantages were the small sample and the speed, quickness, and low cost. Hospital data provided more detailed causes of death and substandard care factors. Community data would require a very large sample in order to achieve greater reliability.