We evaluated a novel on-site antiretroviral therapy programme in a South African correctional facility using routinely collected programme data, from a retrospective cohort of adult inmates starting antiretroviral therapy between 03/2007 and 03/2009 followed-up to 09/2009. We report (1) mortality (using survival analysis); (2) retention to the programme (to 09/2009); and (3) virological suppression at 6 and 12 months (<400 copies/ml) following antiretroviral therapy initiation. In total, 404 started antiretroviral therapy (median age 33 years; 91.3% men; median baseline CD4 count 152 cells/µl [interquartile range 85–225]). Among 299 starting antiretroviral therapy for the first time (antiretroviral therapy-naïve), 23 deaths occurred during 252 person-years (median follow-up 9 months). Mortality rates were 17.2 at 0–6 months (95% confidence interval 10.9–26.9) and 2.8 at >6 months (95% confidence interval 1.1–7.5)/100 person-years; p < 0.001. On 09/2009, 35.6% (144/404) remained in the correctional facility, with 94.4% (136/144) retained to the programme; 38.4% (155/404) were released; and 20.0% (81/404) transferred to another facility. Among antiretroviral therapy-naïve patients in care six and 12 months after antiretroviral therapy initiation, 94.7% (124/131) and 92.5% (74/80) were virologically suppressed, respectively. High early mortality warrants the early identification and management of HIV-positive inmates. The high mobility of inmates necessitates systems for facilitating the continuity of care. Good virological responses and retention supports decentralising HIV care to correctional facilities.