The pattern of change over time of CD4-lymphocyte counts was investigated prospectively in 420 patients infected with HIV-1 (106 women, 314 men; mean age 34.9 [21-70] years) to assess its value in prognosticating progression to AIDS. Only those CD4-lymphocyte values were taken into account which had been measured in the asymptomatic stage of the HIV infection and before introduction of antiviral treatment. An average of 4.5 (range 3-10) measurements per person were available. Mean observation time was 2.4 years, the mean annual CD4-lymphocyte decrease was 48 cells/microliters. 121 patients (28.8%) had a fall in CD4-lymphocyte count which in the regression analysis significantly differed from zero. In this group there occurred 19 progressions to AIDS (15.7%), in contrast to only 20 (6.6%) in the 299 patients with a nonsignificant CD4 fall (P < 0.01). Multivariate analysis with the Cox regression indicated that the annual reduction in the CD4 count and the initial CD4 count were the only values of prognostic significance regarding progression to AIDS. There was no evidence that percentage and absolute CD4-lymphocyte counts had a different predictive value. When the initial CD4 count was high (> 900 cells/microliters), there was on average a steeper reduction in CD4-lymphocyte counts than when it was lower. This findings argues against a linear CD4 fall during the total period of observation. The annual CD4 fall, described with linear regression, is a prognostic criterion on its own for early recognition of those patients at a high risk of progression to AIDS while still in the asymptomatic stage of HIV infection.