This study aimed to identify factors associated with nonadherence during the maintenance phase of highly active antiretroviral therapy (months 12-60) in the Anti PROtease Cohort (APROCO) cohort after correcting for the bias due to missing outcome data. A Heckman 2-stage approach (generalized estimating equations probit model) was used to compare visits with moderate or poor adherence and visits with high adherence. Between months 12 and 60, at least 1 self-reported adherence measure was available for 970 of the 1110 patients with at least 12 months of follow-up (3889 visits with adherence assessments). Adherence was rated as high at 2466 visits, moderate at 1125, and poor at 298. After adjustment for "missingness," moderate and poor adherence were independently associated with age (younger), perceived treatment side effects, dosing frequency different from twice daily, and a protease inhibitor-based regimen. They were also associated with depression and lack of support from the main partner. High adherence was most likely among patients born outside the European Union. A comparison restricted to poor-adherence and high-adherence visits yielded a similar pattern of predictors. Adjusting for missing outcome data changed the predictor set. Reasons for nonadherence are multifactorial. Psychosocial interventions and the selection of the best-tolerated regimens are needed to improve long-term adherence of HIV-infected patients to their lifelong treatment.