E138K, a G→A mutation in HIV-1 reverse transcriptase (RT), is preferentially selected by etravirine (ETR) and rilpivirine over other substitutions at position E138 that offer greater drug resistance. We hypothesized that there was a mutational bias for the E138K substitution and designed an allele-specific PCR to monitor the emergence of E138A/G/K/Q/R/V during ETR selection experiments. We also performed competition experiments using mutated viruses, and quantified the prevalence of E138 minority species in drug-naïve patients. E138K, as well as E138G, consistently emerged first during ETR selection experiments, followed by E138A and E138Q; E138R was never selected. Surprisingly, E138K was identified as a tiny minority in 23% of drug-naïve subtype B patients, a result confirmed by ultradeep sequencing (UDS). This result could reflect a low fitness cost of E138K; however, E138K was one of the least fit substitutions at codon E138, even after taking into account the dNTP pools of the cells used in competition experiments. Further UDS analysis revealed other minority species in a pattern consistent with the mutational bias of HIV RT. There was no evidence of APOBEC3-hypermutation in these selection experiments or in patients. Our results confirm the mutational bias of HIV-1 in patients and highlight the importance of G→A mutations in HIV-1 drug resistance evolution.