The clinical utility of the World Health Organization (WHO) clinical case definition (CCD) of acquired immune deficiency syndrome (AIDS) in Africa, several proposed modifications of the WHO CCD, and two proposed screening algorithms for human immunodeficiency virus (HIV) infection were examined in adult medical inpatients in Dar es Salaam, Tanzania. Sensitivity, specificity, and positive and negative predictive values were determined for the CCDs and screening algorithms. Multivariable analysis identified factors with high accuracy for HIV infection. Of 223 patients enrolled in the study, 95 were seropositive for HIV infection. The WHO CCD and the modified CCDs had low sensitivities (14.7-32.6%) but high specificities (95.3-99.2%) and positive predictive values (83.8-94.7%). The screening algorithms had moderate sensitivities (66.3-77.9%) and poor specificities (46.1-79.7%). Multivariable analysis consistently identified oral candidiasis and lymphadenopathy as the best predictors of HIV infection. Although patients with asymptomatic or early HIV infection may be missed by clinical criteria, in a high prevalence population, AIDS may be diagnosed accurately clinically because of the effect of prevalence on the positive predictive values of the CCDs. Furthermore, selection of patients for HIV serologic testing may be guided by simple combinations of clinical features.