Adherence to highly active antiretroviral treatment (HAART) has been associated with increased survival rates and decreased drug resistance in various settings. There is growing concern that loss to follow-up will increase and adherence rates will decrease as HAART programs are expanded in resource-limited settings. In Central Mozambique, an innovative program was implemented, using community-based (trained community activists) and self-selected (family members or friends) "treatment partners" to provide psycho-social support to patients on HAART. We calculated adherence rates based on pharmacy records for all patients who refilled their medication for at least six consecutive months between September 2004 and June 2006. Medical charts were reviewed for a subset of 375 patients having high (> or =90%) adherence and 59 patients having low (<90%) adherence. Multivariate logistic regression analysis assessed the association between the type of treatment partner used and adherence to HAART. A total of 305 patients (70%) had self-selected treatment partners, 121 (28%) had community-based treatment partners, and 8 (2%) had no treatment partner. In adjusted analysis, patients who had no treatment partner were more likely to have low adherence (OR 9.47; 95% confidence interval 2.37-37.86 compared to self-selected treatment partner). Patients with community-based treatment partners did not have significantly lower adherence than patients with self-selected treatment partners. While it cannot be determined from these data which aspects or types of peer support are most effective in maintaining adherence, it appears that peer support was beneficial to this study population. While the study results are not directly applicable to other populations, other HAART programs should consider the potential benefit of providing treatment support to patients.