Perinatal mortality is a sensitive indicator of health status of a community and is also highly amenable to intervention. The causes of perinatal deaths in developing countries are often difficult to establish. Verbal autopsy has been used in several countries for children and adults, but seldom for perinatal cause.To establish the cause of perinatal deaths using verbal autopsy.Community-based cross-sectional, retrospective study to identify perinatal death over a one year period from July 1996-June 1997. Comparison was made with hospital records. An algorithm of signs and symptoms was used by trained personnel to identify the cause of perinatal death. The duration of collection of data was six months (August 1996-January 1997).Hai district of Kilimanjaro region in Tanzania.All perinatal deaths within one year.The perinatal mortality was 58 per 1000 (121 deaths and 2088 live births). Verbal autopsy could establish the cause of death in 105 of the 121 deaths. Hospital records showed 79 deaths indicating that 42 deaths probably occurred at home. Among the 79 available hospital records, the cause of death could be established in only 30 (38%). The causes of postnatal death were compared between the verbal autopsy and hospital records. There was a good correlation between the same, however only 18 records were available from hospital among the total 31 postnatal deaths. The specificity of determining cause of death using verbal autopsy was 100% and sensitivity 61%.The commonest causes of perinatal deaths were related to obstetric care, therefore interventions to curb perinatal mortality should be directed to improvement of obstetric care. Verbal autopsy is a simpler and more sensitive tool in establishing the cause of perinatal death than hospital records in a rural district of Tanzania. Large-scale studies are needed to validate this.