BACKGROUND: The scale-up of antiretroviral treatment (ART) services in resource-limited settings requires a programmatic model to deliver care to large numbers of people. Understanding the determinants of key outcome measures--including death and nondeath losses--would assist in program evaluation and development. METHODS: Between September 2002 and August 2005, all in-program (pretreatment and on-treatment) deaths and nondeath losses were prospectively ascertained among treatment-naive adults (n=1235) who were enrolled in a community-based ART program in South Africa. RESULTS: At study censorship, 927 patients had initiated ART after a median of 34 days after enrollment in the program. One hundred twenty-one (9.8%) patients died. Mortality rates were 33.3 (95% CI, 25.5-43.0), 19.1 (95% CI, 14.4-25.2), and 2.9 (95% CI, 1.8-4.8) deaths/100 person-years in the pretreatment interval, during the first 4 months of ART (early deaths), and after 4 months of ART (late deaths), respectively. Pretreatment and early treatment deaths together accounted for 87% of deaths, and were independently associated with advanced immunodeficiency at enrollment. Late deaths were comparatively few and were only associated with the response to ART at 4 months. Nondeath program losses (loss to follow-up, 2.3%; transfer-out, 1.9%; relocation, 0.7%) were not associated with immune status and were evenly distributed during the study period. CONCLUSIONS: Loss to follow-up and late mortality rates were low, reflecting good cohort retention and treatment response. However, the extremely high pretreatment and early mortality rates indicate that patients are enrolling in ART programs with far too advanced immunodeficiency. Causes of late access to the ART program, such as delays in health care access, health system delays, or inappropriate treatment criteria, need to be addressed.