To analyze how demographic, clinical, and laboratory characteristics influence the risk of tuberculosis in human immunodeficiency virus (HIV)-infected individuals; to examine the incidence of tuberculosis associated with change in skin test responsiveness in HIV-infected, tuberculin-negative, nonanergic individuals. Multicenter cohort study. Twenty-three infectious disease units in public hospitals in Italy. A consecutive sample of 3397 HIV-infected subjects were considered for entry in the study. Of these, 2695 who were followed up for at least 4 weeks were enrolled in the study; 739 subjects (27.4%) were unavailable for follow-up. The median duration of follow-up was 91 weeks. Culture-proven tuberculosis. Eighty-three episodes of tuberculosis were observed. Incidence rates of tuberculosis were 5.42 per 100 person-years among tuberculin-positive subjects, 3.00 per 100 person-years among anergic subjects, and 0.45 per 100 person-years among tuberculin-negative nonanergic subjects. In multivariate analysis, being tuberculin-positive (hazard ratio [HR], 9.94; 95% confidence interval [CI], 3.84 to 25.72) or anergic (HR, 3.35; 95% CI, 1.40 to 8.00), or having a CD4+ lymphocyte count less than 0.20 x 10(9)/L (HR, 4.87; 95% CI, 2.35 to 10.11) or between 0.20 and 0.35 x 10(9)/L (HR, 2.35; 95% CI, 1.09 to 5.05) were statistically significantly associated with the risk of tuberculosis. Incidence of tuberculosis increased with decreasing levels of CD4+ lymphocytes in the three groups of subjects with different skin test responsiveness. Skin tests were repeated 1 year after enrollment in 604 tuberculin-negative nonanergic subjects; three case of tuberculosis were observed among the 13 subjects who converted to tuberculin reactivity. Risk of tuberculosis in HIV-infected persons can be more precisely quantified by jointly considering skin test reactivity and CD4+ lymphocyte count. Periodic skin tests in tuberculin-negative nonanergic individuals can be useful in identifying individuals at high risk of active tuberculosis.