The effectiveness of the South African government's expanding antiretroviral treatment program is unknown. Observational studies of treatment effectiveness are prone to selection bias, rarely compare patients receiving antiretroviral treatment with similar patients not receiving antiretroviral treatment, and underestimate mortality rates unless patients are actively followed up. We followed up 14 267 patients in the Public Sector Anti-Retroviral Treatment project in Free State, South Africa, for up to 20 months after enrollment. A total of 3619 patients received highly active triple antiretroviral treatment (HAART) for up to 19 months (median, 6 months; interquartile range, 3-9 months) after enrollment. Patients' clinical data were linked with the national mortality register. Marginal structural regression models adjusted for baseline and time-varying covariates. Of 4570 patients followed up for at least 1 year, 53.2% died. Eighty-seven percent of patients who died had not received HAART. HAART was associated with lower mortality (hazard ratio, 0.14; 95% confidence interval [CI], 0.11-0.18) and with the presence of tuberculosis (hazard ratio, 0.61; 95% CI, 0.46-0.81) after adjusting for age, sex, weight, clinic, district, CD4 cell count, cotrimoxazole therapy, tuberculosis at baseline, and previous antiretroviral therapy. Cotrimoxazole therapy was associated with lower mortality (hazard ratio, 0.37; 95% CI, 0.32-0.42). Each month of HAART was associated with an increase in CD4 cell count of 15.1 cells/microL (95% CI, 14.7-15.5 cells/microL) and with an increase in body weight of 602 g (95% CI, 548-658 g). HAART provided through these South African government health services seems as effective as that provided in high-income countries. Delays starting HAART contributed to high mortality rates. Faster expansion and timely commencement of HAART are needed.