Transforming natural units of laboratory markers of HIV disease may improve their ability to predict clinical outcomes. The authors examined this issue within a prospective study of 394 patients enrolled in the Swiss HIV Cohort Study (SHCS) between 1991 and 1993. Baseline predictors included CD4+ and CD8+ cell counts, HIV RNA levels, β2-microglobulin, and age. Outcomes were death and clinical progression. A range of power transformations was applied to each predictor, and the goodness-of-fit of the corresponding proportional hazards models was assessed. The prognostic value of all laboratory variables could be improved on by power transformations. To predict either outcome variable, the “best” transformation of HIV RNA copies and CD8+ cell counts was the logarithm; for β2-microglobulin, it was power -2. For CD4+ cell counts, the best transformation depended on the outcome variable: it was power 0.2 when predicting survival, and power 0.4 when predicting clinical progression. The single best predictor variable was the ratio of HIV RNA copies per CD4+ cell, for both death (logarithmic transformation) and clinical progression (power -0.1 transformation). Natural units of laboratory variables are not optimal for the prediction of clinical events in HIV-infected patients. Which transformation is best depends on the predictor under consideration.