A prospective multicenter cohort study comprising 1,171 individuals who were seropositive for human immunodeficiency virus (HIV) but did not have AIDS at the time of enrollment and 182 HIV-seronegative controls, was studied by means of routine induced-sputum analysis in an attempt to detect occult tuberculosis or Pneumocystis carinii pneumonia. One occult case of tuberculosis was discovered upon the patient's enrollment (at baseline); none were discovered during follow-up. Two additional Mycobacterium tuberculosis isolates were recovered (one at baseline, one during follow-up) from subjects with symptoms or abnormalities evident on chest roentgenograms. Three specimens were false-positive (one for M. tuberculosis, two for P. carinii). Five pathogenic nontuberculous mycobacteria isolates were recovered during follow-up. Nonpathogenic, nontuberculous mycobacteria were recovered from 51 (4.6%) of 1,113 baseline specimens and 56 (3.7%) of 1,518 follow-up specimens, primarily at a center where the water supply was contaminated. We conclude that routine induced-sputum analysis is not an effective strategy for screening HIV-infected asymptomatic subjects for tuberculosis or P. carinii pneumonia before the onset of clinically recognizable disease activity.