To determine the relationships between individuals' baseline T-cell subsets, their rates of change with time, and AIDS-free survival time, data were collected at 6-monthly intervals from 379 HIV-seropositive homosexual Sydney men, of whom 31 developed AIDS during the 3-year observation period. Both CD4% and rate of change of CD4% in an individual had significant prognostic value in determining AIDS-free survival time. Compared with subjects whose CD4% remained stable, subjects whose CD4% dropped by 7% or more in a year had a relative hazard of 35.1 (95% confidence interval = 11.7–105.6,P< 0.001) of developing AIDS. Increasing CD4% had a significant protective effect, reducing the risk of developing AIDS. CD4%, CD4 cell count and CD4:CD8 ratios showed steeper declines in subjects who were later diagnosed with AIDS than in those who remained AIDS-free. The rates of immunological change in AIDS-free seroconverters and seropositives were similar, despite indeterminate differences in durations of infections. In the multivariate Cox regression analysis, baseline CD4%, the rate of change of CD4%, and baseline lymphocyte count were associated with AIDS-free survival time. Baseline CD4% had greater prognostic value than baseline CD4 cell count. Baseline CD8%, baseline CD8 count, their rates of change and their mean square errors were not independently significant in this analysis. These findings are important for clinicians monitoring HIV infection in an individual and for entry criteria and monitoring procedures in clinical trials. They also have implications for resource-poor settings; prognosis based on CD4% can be made with a flow cytometer without a full blood count. Data were collected before zidovudine was available and thus reflect the natural history of HIV infection.