To examine predictors of magnitude of CD4+ response to treatment of human immunodeficiency virus (HIV) infection with zidovudine. This was a post hoc analysis of randomized placebo-controlled clinical trial in a multicenter trial, 1423 asymptomatic HIV-positive subjects with CD4+ cell counts less than 500 mm-3 were given 500 mg/day zidovudine, 1500 mg/day zidovudine, or placebo. The main outcome measure was change in the CD4+ cell counts over time. This study suggests that earlier treatment with zidovudine results in a larger increment in the CD4+ cell count. In addition, the increment in CD4+ cell count is very long lived. However, drug exposure was not found to be a predictor of response to treatment in the dose range studied. A parametric model of disease progression can be estimated with use of data collected in a conventionally designed study. These parametric models may provide insight into the optimal use of drugs. This model suggests that zidovudine does not change the underlying course of HIV infection but simply delays the time course. The model also suggests that the magnitude of this delay is larger when treatment is begun earlier in the course of the disease.