In Mozambique, the evaluation of retention in HIV care and ART programmes is limited. To assess rate and predictors of attrition (no retention in care) and HAART effectiveness in HIV-1 infected patients who pay for medication and laboratory testing in Mozambique, we conducted a multicenter survey of HIV-1-infected patients who started HAART during 2002-2006. Cox proportional hazard models were used to assess risk of attrition and of therapy failure. Overall, 142 patients from 16 healthcare centers located in the capital city Maputo were followed-up for 22.2 months (12.1-46.7). The retention rate was 75%, 48% and 37% after one, two and three years, respectively. Risk of attrition was lower in patients with higher baseline CD4 count (P = 0.022) and attending healthcare center 1 (HCC1) (P = 0.013). The proportion of individuals with CD4 count ≤ 200 cells/µL was 55% (78/142) at baseline and decreased to 6% (3/52) at 36 months. Among the patients with available VL, 86% (64/74) achieved undetectable VL levels. The rate of immunologic failure was 17.2% (95% CI: 12.6-22.9) per 100 person-years. Risk of failure was associated to higher baseline CD4 count (P = 0.002), likely reflecting low adherence levels, and decreased with baseline VL ≥ 10,000 copies/mL (P = 0.033). These results suggest that HAART can be effective in HIV-1 infected patients from Mozambique that pay for their medication and laboratory testing. Further studies are required to identify the causes for low retention rates in patients with low CD4 counts and to better understand the association between healthcare setting and attrition rate.