OBJECTIVE: To assess the incidence, magnitude and factors associated with the first episode of non-adherence for 12 months after the first antiretroviral prescription. DESIGN: A prospective study of HIV-infected patients receiving their first antiretroviral prescription in public referral centers, Belo Horizonte, Brazil. Baseline assessment occurred at the moment of the first prescription and follow-up visits at the first, fourth and seventh month, from May 2001 to May 2003. METHODS: Non-adherence was self-reported and defined as the intake of less than 95% of the prescribed doses for 3 days before the follow-up interviews. Cumulative and person-time incidence were estimated and Cox's proportional model was used to assess the relative hazard (RH) of non-adherence with 95% confidence interval for both univariate and multivariate analysis. RESULTS: Among 306 patients, the cumulative incidence of non-adherence was 36.9% (incidence rate 0.21/100 person-days). Multivariate analysis (P < 0.05) showed that unemployment (RH = 2.17), alcohol use (RH = 2.27), self-report of three or more adverse reactions (RH = 1.64), number of pills per day (RH = 2.04), switch in antiretroviral regimen (RH = 2.72), and a longer time between the HIV test result and the first antiretroviral prescription (RH = 2.27) were associated with an increased risk of non-adherence, whereas the use of more than one health service indicated a negative association (RH = 0.54). CONCLUSION: The current analysis has pointed out the importance of clinical and health service characteristics as potential indicators of non-adherence after initiating therapy. Early assessment and intervention strategies should be priorities in these AIDS public referral centres. Feasible and reliable indicators for the routine monitoring of adherence should be incorporated in clinical practice.