dIdentifying human immunodeficiency virus (HIV) immune escape mutations has implications for understanding the impact of host immunity on pathogen evolution and guiding the choice of vaccine antigens. One means of identifying cytotoxic-T-lymphocyte (CTL) escape mutations is to search for statistical associations between mutations and host human leukocyte antigen (HLA) class I alleles at the population level. The impact of evolutionary rates on the strength of such associations is not well defined. Here, we address this topic using a mathematical model of within-host evolution and between-host transmission of CTL escape mutants that predicts the prevalence of escape mutants at the population level. We ask how the rates at which an escape mutation emerges in a host who bears the restricting HLA and reverts when transmitted to a host who does not bear the HLA affect the strength of an association. We consider the impact of these factors when using a standard statistical method to test for an association and when using an adaptation of that method that corrects for phylogenetic relationships. We show that with both methods, the average sample size required to identify an escape mutation is smaller if the mutation escapes and reverts quickly. Thus, escape mutations identified as HLA associated systematically favor those that escape and revert rapidly. We also present expressions that can be used to infer escape and reversion rates from cross-sectional escape prevalence data.T he human leukocyte antigen (HLA)-restricted cytotoxic T-lymphocyte (CTL) immune response is thought to make a significant contribution to the control of human immunodeficiency virus (HIV) (4,5,13,27). A deeper understanding of the CTL response is important to the development of an HIV vaccine. One way to study the CTL response is by investigating the way in which HIV is evolving escape mutants-viral strains that evade recognition by CTLs. Evidence to support the evolution of CTL escape mutants has been observed both within individuals (4,21,28,36,39) and at the population level (3,9,17,23,25,33,38). At the population level, evidence has been found in the form of statistical associations between certain HLA class I alleles-the human genetic determinants of CTL responses-and certain mutations away from the sample/subtype consensus in the HIV genome (3, 9, 33). The patterns emerge because of heterogeneity in HLA alleles among the population. Individuals who share HLA alleles tend to target the same viral antigens (called CTL epitopes) and therefore drive the same escape mutations. Escape mutations have been shown to revert to the wild-type form (e.g., back to the subtype consensus) following transmission to hosts who do not bear the selecting HLA (28, 29). The combination of these two factors-escape in "HLA-matched" hosts and reversion in "HLAmismatched" hosts-means that although viral mutants can be transmitted between individuals, any particular escape mutation should be more prevalent in HLA-matched than in HLA-mismatched hosts. Simple statistic...