This is a retrospective descriptive study examining medical records of notified cases of tuberculosis in Birmingham, UK, who died before completion of treatment, plus notified cases identified at postmortem. During the study period, 1989-1995, there were a total of 2088 notifications of tuberculosis of which 75 (3.6%) died. This case-fatality rate is lower than that reported from earlier studies, a finding which may be due to improved ascertainment, earlier diagnosis and improved treatment in recent years. Forty-five deaths (60%) were in males and 30 (40%) in females. The median age at death was 66 y. Forty-five deaths (60%) were among the white ethnic group, 21 (28%) among persons of Indian sub-continent origin, 4 (5.3%) among Black Caribbeans and 5 (6.6%) among the other ethnic groups. Case-fatality rates were significantly higher (P < 0.01) in caucasians (9.4%) than in Asians (1.5%) and only half of this difference was explained by the age of those affected. Twenty-two (29%) cases were confirmed by a positive sputum smear and a further 23 (31%) by a positive sputum culture. Eighteen (24%) cases also had histological confirmation. Twelve (16%) cases were identified on autopsy. The case fatality rate in respiratory disease was significantly higher (RR = 1.19, P > 0.05) than in non-respiratory disease. As delay in diagnosis is likely to be the main contributing factor leading to death, a high index of suspicion of tuberculosis is needed when investigating elderly patients with general chronic illness, especially if there are prolonged respiratory symptoms. Early diagnosis will reduce mortality as the disease rapidly responds to treatment.