The emergence of human immunodeficiency virus (HIV) drug-resistant mutations during antiretroviral therapy is explained by either the preexistence of low-frequency-resistant strains before the start of therapy or by the selection of unsuppressed resistant strains during therapy. We used pairwise and maximum likelihood analyses of the ratio of nonsynonymous to synonymous substitutions per site (d(N)/d(S)) to study the extent of positive selection in the reverse transcriptase (RT) gene of HIV-1 from multiple data sets of drug-treated (117 sequences) and drug-naive patients (270 sequences). In the pairwise analysis, evidence for positive selection (d(N)/d(S) > 1) was only found in drug-treated individuals and in codons conferring drug resistance. By the maximum likelihood method, a positive selection at codons conferring drug resistance was only observed in patients receiving therapy, and although positive selection was detected in drug-naive patients, this was always at codons unrelated to drug resistance. We therefore document a striking difference in the process of allele fixation in drug resistance codons (RC) between populations of HIV-1-infected individuals naive to treatment and those receiving therapy. Furthermore, although mutations associated with drug resistance are sometimes found in drug-naive patients, we suggest that these are fixed because of linkage to sites experiencing immune escape. Finally, we show that compensatory changes are likely to be important in the development of HIV drug resistance by counter-acting the deleterious effects normally associated with drug resistance mutations.