Evaluation of disease outcome is central to the assessment of tuberculosis (TB) control programmes. In the study reported in this article we examined the factors influencing the measurement of outcome, survival rates during and after treatment, smear conversion rates, and relapse rates for patients diagnosed with TB in a rural area of Malawi between 1986 and mid-1994. Patients with less certain diagnoses of TB were more likely to die than those with confirmed TB, both among those who were seropositive and those who were seronegative to human immunodeficiency virus (HIV). The mortality rate among smear-positive patients with a separate culture-positive specimen was half that of patients with no such diagnostic confirmation. Patients not registered by the Ministry of Health had much higher mortality and default rates than did registered patients. Among smear-positive patients, HIV serostatus was the most important influence on mortality both during and after treatment (crude hazard ratios (95% confidence intervals) = 5.6 (3.0-10) and 7.7 (3.4-17), resp.), but HIV serostatus did not influence smear conversion rates. The initial degree of smear positivity influenced smear conversion rates, but not mortality rates. No significant predictors of relapse were identified. Unless considerable care is taken to include all TB patients, and to exclude nontuberculous patients, recorded TB outcome statistics are difficult to interpret and may be misleading. In populations with high rates of HIV infection, TB target cure rates of 85% are unrealistic. When new interventions are assessed it cannot be assumed that factors which influence the smear conversion rate will also influence the mortality rate.