The perinatal mortality rate (PNMR) is a key health status indicator. It is multifactorial in aetiology and is significantly influenced by the quality of health care. While there is an ethical imperative to act to improve quality of care when deficiencies are apparent, the lack of controls - when an intervention is applied to an entire service - makes it difficult to infer a causal relationship between the intervention and any subsequent change in PNMR. However, by specifically measuring avoidable perinatal deaths (those due to error or omission on the part of the health service), this limitation is partially overcome, and the impact of the intervention can be more rigorously evaluated. This paper reports the impact of perinatal audit in a rural African health district between 1991 and 1995. A total of 21,112 consecutive births were studied: the average number of deliveries increased by 31% from 325 to 424 per month. The PNMR (birth weight ≤ 1000 g) in 1991 was 27/1000, increased to 42/1000 in 1992, and fell steadily to 26/1000 in 1995 (40% reduction; p = 0.002). The proportion of avoidable deaths fell from 19% in 1991 to zero in the second half of 1995 (p = 0.0008). While factors associated with perinatal mortality are many, complex, and interrelated, this report suggests that mortality can be reduced significantly in resource-poor settings by improving quality of health care. Including the measurement of avoidable deaths in perinatal audit allows the impact of interventions to be more rigorously assessed than by simply measuring the PNMR. This study determines the perinatal mortality rates (PNMR) in Hlabisa Maternity Hospital and eight village clinics in KwaZulu/Natal, South Africa, during May 1991 to December 1995. The PNMR is the number of stillbirths and the number of deaths before discharge per 1000 total births. Avoidable perinatal death is death determined to be directly due to an error or omission on the part of the health service. Interventions aimed to increase the quality of care and reduce avoidable perinatal mortality. Interventions included structural and functional changes in the maternity services throughout the district, using protocols for care, and conducting in-service training. Hlabisa Hospital and the eight clinics saw 21,112 births during the study's time period. The average monthly number of deliveries increased from 325 in 1991 to 424 in 1995. 653 perinatal deaths occurred. The PNMR for the 8-month period in 1991 was 27/1000. PNMR was 36/1000 in the first half of 1992, 42/1000 in the second half, and declined to 26/1000 thereafter. 19% of deaths were avoidable in 1991; zero deaths were avoidable in 1995. The proportion of births in clinics was 35% of the total births throughout the time period. Perinatal deaths in clinics declined from 17% in 1991 to 6.3% in the second half of 1995. This study indicates that quality of care improvements are possible, despite increased workload. Two elements were considered crucial in the early stages of the intervention: clinic staff's adoption of the concept of district-wide services and opportunity to refer any cases to neighboring hospitals. Avoidable deaths were reduced by 61%, despite a 16% increase in workload. The initial rise in PNMR is attributed to the shift to managing high-risk cases in hospitals. PNMR decline is attributed to the continuous and rigorous audit of services and to the use of the concept of avoidable deaths for monitoring quality of care.