The aim of this study was to clinically validate a heminested polymerase chain reaction (PCR) method, based on the IS6110 insertion segment of Mycobacterium tuberculosis complex, for the diagnosis of tuberculosis. Samples of pulmonary, extrapulmonary and blood origin were collected prospectively from 331 patients. All samples were processed to detect acid-fast bacilli by direct stain, culture and PCR. The gold standard comparison was a clinically based final case definition of tuberculosis corresponding to group 3 of the American Thoracic Society's classification system. The sensitivities of stain, culture and PCR were 41%, 65% and 59%, respectively. Overall specificity exceeded 97% for all techniques. The combination of PCR and direct stain achieved a sensitivity similar to that of culture alone. The PCR method detected 74 of 95 (78%) culture-positive results. In a hospital setting, PCR could be a useful, reliable tool for diagnosis of tuberculosis and may be introduced as a complementary routine diagnostic laboratory method.