Objective: To compare the prevalence of resistance by adherence level in patients treated with non-nucleoside reverse transcriptase inhibitors (NNRTI) or protease inhibitors (PI). Also to examine the mechanism of differential class-specific adherence-resistance relationships, focusing on the patient-derived capacity of wild-type and drug-resistant recombinant variants to replicate in vitro in the presence of variable drug levels. Methods: Participants received unannounced pill count measures to assess adherence, viral load monitoring, and genotypic resistance testing. The replicative capacity of drug-susceptible and drug-resistant recombinants was determined using a single-cycle recombinant phenotypic susceptibility assay. Drug exposure was estimated using population-averaged pharmacological measurements adjusted by participant-specific levels of adherence. Results: In the NNRTI-treated group, 69% had resistance at 0-48% adherence compared to 13% at 95-100% (P = 0.01 ). PI resistance was less common than NNRTI resistance at 0-48% adherence (69% versus 23%; P = 0.01). In multivariate analysis, the odds for PI resistance increased (P = 0.03) while the odds for NNRTI resistance decreased (P = 0.04) with improving adherence. Individuals with drug-resistant variants were more likely to have levels of drug exposure where the resistant variant was more fit than the drug-susceptible variant in vitro, while those with drug-susceptible virus were more likely to have levels of drug exposure where the drug-susceptible virus was more fit than the drug-resistant variant (P = 0.005). Conclusions: NNRTI resistance was more common than PI resistance at low levels of adherence. Class-specific adherence-resistance relationships are associated with the relative replicative capacity of drug-resistant versus wild-type variants to replicate in the presence of clinically relevant drug levels. © 2006 Lippincott Williams & Wilkins.