Information from routine and sentinel surveillance was used to monitor the HIV/AIDS epidemic in KwaZulu-Natal, South Africa between 1991 and 1997. Comparisons were made between data obtained from (1) sentinel surveillance for antenatal HIV infection, pulmonary tuberculosis (PTB), and AIDS in a single health district and (2) province-wide sentinel surveillance for antenatal HIV infection, legally required notification of cases of PTB, and voluntary notification of AIDS cases. HIV prevalence among antenatal clinic attenders in the sentinel district rose rapidly and at similar rates to provincial figures: 4.2% vs 4.8% in 1992 to 25.9% vs 26.9% in 1997. PTB incidence increased four-fold in the sentinel district over the study period, whereas provincial PTB figures from passive surveillance fluctuated widely and showed no clear increase (Chi-square for trend 425.5, P<0.00001). AIDS incidence in the sentinel district increased dramatically while provincial data from the voluntary reporting system showed a less consistent and much slower rise (Chi-square for trend 9.07, P=0.003). Incidence of AIDS in 1997 was estimated as 437/10(5) in the sentinel district compared to 32/10(5) in the provincial figures. Routine disease notification and voluntary reporting systems are likely to underestimate the impact of the HIV/AIDS epidemic in resource-poor settings. Sentinel surveillance at representative sites should be developed to validate or replace passive surveillance systems.