Sentinel surveillance surveys combined with epidemiological/demographic models have been used in monitoring trends in mortality due to HIV in South Africa as in most other countries. The potential biases in such surveys have been highlighted in recent studies. South Africas HIV prevalence (29.5% among pregnant women attending antenatal clinics in 2004) is one of the highest in the world. National HIV prevalence from population-based survey has been carried out twice in South Africa, but is of limited use as a surveillance system independent of antenatal data. The use of vital registration in monitoring HIV related mortality has been largely unexplored. This study utilises data on registered deaths for the period 1997-2001 to estimate trends in mortality and TB/HIV related death rates in South Africa. The analysis utilizes the Growth Balance method to estimate the completeness of the death records from the registration system, and hence an adjustment factor for the registered deaths including TB/HIV related deaths in the period. The results indicate that coverage of registration of deaths in South Africa has increased over time but the completeness of registration of female deaths lags behind completeness of registration of male deaths. The absolute number of deaths is estimated to have increased from about 400,000 in 1997 to about 565,000 in 2002. Crude HIV related plus TB-AIDS death rate is estimated to have increased from about 9.6 per 10,000 in 1997 to about 19.1 per 10,000 in 2002. The increase in the number of deaths could not be attributed to population growth.