Human immunodeficiency virus (HIV-1) can rapidly evolve due to selection pressures exerted by HIV-specific immune responses, antiviral agents, and to allow the virus to establish infection in different compartments in the body. Statistical models applied to HIV-1 sequence data can help to elucidate the nature of these selection pressures through comparisons of non-synonymous (or amino acid changing) and synonymous (or amino acid preserving) substitution rates. These models also need to take into account the non-independence of sequences due to their shared evolutionary history. We review how we have developed these methods and have applied them to characterize the evolution of HIV-1 in vivo. To illustrate our methods, we present an analysis of compartment-specific evolution of HIV-1 env in blood and cerebrospinal fluid and of site-to-site variation in the gag gene of subtype C HIV-1.