BACKGROUND: HAART efficacy was evaluated in a real-life setting in Phnom Penh (Médecins Sans Frontières programme) among severely immuno-compromised patients. METHODS: Factors associated with mortality and immune reconstitution were identified using Cox proportional hazards and logistic regression models, respectively. RESULTS: From July 2001 to April 2005, 1735 patients initiated HAART, with median CD4 cell count of 20 (inter-quartile range, 6-78) cells/microl. Mortality at 2 years increased as the CD4 cell count at HAART initiation decreased, (4.4, 4.5, 7.5 and 24.7% in patients with CD4 cell count > 100, 51-100, 21-50 and < or = 20 cells/microl, respectively; P < 10). Cotrimoxazole and fluconazole prophylaxis were protective against mortality as long as CD4 cell counts remained < or = 200 and < or = 100 cells/microl, respectively. The proportion of patients with successful immune reconstitution (CD4 cell gain > 100 cells/microl at 6 months) was 46.3%; it was lower in patients with previous ART exposure [odds ratio (OR), 0.16; 95% confidence interval (CI), 0.05-0.45] and patients developing a new opportunistic infection/immune reconstitution infection syndromes (OR, 0.71; 95% CI, 0.52-0.98). Similar efficacy was found between the stavudine-lamivudine-nevirapine fixed dose combination and the combination stavudine-lamivudine-efavirenz in terms of mortality and successful immune reconstitution. No surrogate markers for CD4 cell change could be identified among total lymphocyte count, haemoglobin, weight and body mass index. CONCLUSION: Although CD4 cell count-stratified mortality rates were similar to those observed in industrialized countries for patients with CD4 cell count > 50 cells/microl, patients with CD4 cell count < or = 20 cells/microl posed a real challenge to clinicians. Widespread voluntary HIV testing and counselling should be encouraged to allow HAART initiation before the development of severe immuno-suppression.