We evaluated the age profile and chest radiographic abnormalities in 158 patients from Bujumbura, Burundi, with new-onset intrathoracic tuberculosis (pulmonary, pleural, or hilar/mediastinal adenopathy), to identify features that were associated with and would allow prediction of HIV seropositivity or seronegativity. Using agreed-upon criteria and prepared reporting forms, initial chest radiographs were reviewed by three readers, first independently and then at a consensus conference. Of the 158 patients, 105 (66%) were HIV seropositive and 53 patients were seronegative. Seropositive subjects (mean age, 35.8 yr) were older (p = 0.001) than seronegative subjects (mean age, 29.4 yr). Significant or borderline differences between HIV-seropositive and -seronegative patients included the frequency of small nodular lesions (p = 0.03), upper lobe cavitation (p = 0.05), and lymphadenopathy (p = 0.12), and the location of parenchymal abnormalities (p = 0.0006). Stepwise logistic regression revealed four important variables: age, small lesions, location, and lymphadenopathy; these were then used to derive an equation to calculate the probability that a given tuberculosis patient was HIV seropositive. Our mathematical model fit the observed data and the equation predicted serologic findings reasonably well. We conclude that it is possible to determine with useful probability a Burundian tuberculosis patient's HIV serologic status.