Data from the Multicenter AIDS Cohort Study (MACS) of 1637 gay men, recruited in 1984 and 1985 in Los Angeles and followed at 6-monthly intervals, are used to develop a computer simulation model of the typical pattern of CD4 T-cell number changes in HIV-infected AIDS-free subjects. The empirical model incorporates the following features: (1) within-person and between-person variability in CD4 measurements; (2) variation in the rates of decline of CD4 values; (3) variation in the level of CD4 at which clinical AIDS is diagnosed, and (4) greater absolute variation in CD4 values in men with high CD4 levels compared with men with low CD4 values. Three applications of the model to assist in the design and interpretation of clinical trials are given. Further applications to clinical trials and to estimate the current and future spectrum of HIV-mediated immunological disease in the USA, as measured by the CD4 values, are discussed.