OBJECTIVES: We assessed retention and predictors of attrition (recorded death or loss to follow-up) in antiretroviral treatment (ART) clinics in Tanzania, Uganda and Zambia. METHODS: We conducted a retrospective cohort study among adults (≥18 years) starting ART during 2003-2010. We purposefully selected six health facilities per country and randomly selected 250 patients from each facility. Patients who visited clinics at least once during the 90 days before data abstraction were defined as retained. Data on individual and programme level risk factors for attrition were obtained through chart review and clinic manager interviews. Kaplan-Meier curves for retention across sites were created. Predictors of attrition were assessed using a multivariable Cox-proportional hazards model, adjusted for site-level clustering. RESULTS: From 17 facilities, 4147 patients were included. Retention ranged from 52.0% to 96.2% at 1 year to 25.8%-90.4% at 4 years. Multivariable analysis of ART initiation characteristics found the following independent risk factors for attrition: younger age [adjusted hazard ratio (aHR) and 95% confidence interval (95%CI) = 1.30 (1.14-1.47)], WHO stage 4 ([aHR (95% CI): 1.56 (1.29-1.88)], >10% bodyweight loss [aHR (95%CI) = 1.17 (1.00-1.38)], poor functional status [ambulatory aHR (95%CI) = 1.29 (1.09-1.54); bedridden aHR1.54 (1.15-2.07)], and increasing years of clinic operation prior to ART initiation in government facilities [aHR (95%CI) = 1.17 (1.10-1.23)]. Patients with higher CD4 cell count were less likely to experience attrition [aHR (95%CI) = 0.88 (0.78-1.00)] for every log (tenfold) increase. Sites offering community ART dispensing [aHR (95%CI) = 0.55 (0.30-1.01) for women; 0.40 (0.21-0.75) for men] had significantly less attrition. CONCLUSIONS: Patient retention to an individual programme worsened over time especially among males, younger persons and those with poor clinical indicators. Community ART drug dispensing programmes could improve retention.