To investigate Mycobacterium vaccae immunotherapy in the treatment of human tuberculosis and to assess longer-term outcomes following treatment for tuberculosis patients, a double-blind placebo-controlled Phase-2 clinical trial was set up in the Mseleni and Manguzi health wards in north-eastern KwaZulu, South Africa. In 1991-93, 204 patients admitted with clinical tuberculosis to the 2 hospitals were allocated to receive single intradermal doses of 0.1 mL M. vaccae NCTC 11659 or 0.1 mL tetanus toxoid alongside standard 6-months chemotherapy with rifampicin, isoniazid, pyrazinamide and ethambutol. The main outcome measures were sputum bacteriology culture conversion to negativity, clinical assessment, weight gain, erythrocyte sedimentation rate and chest radiography. Patients were followed-up after 4 years to determine their health status. M. vaccae cases gained weight more quickly during the first 8 weeks compared with 'placebo' patients. Regression analysis found a synergistic relationship between BCG positive scar status and M. vaccae-induced weight gain. No further difference was found between treatment groups. The bacterial conversion rate to negativity at 2 months was much lower than expected (44.8% active, 38.8% placebo). Mortality was considerably higher than expected after treatment (7.1% each group) and after 4 years (25.8% active, 21.0% placebo; death from tuberculosis 14.5% and 16.1%, respectively). Immune sensitization to environmental mycobacteria may explain the geographical variability of M. vaccae efficacy, as occurs with BCG vaccination and occurred with Koch's tuberculins of the late 19th century. Multiple doses of M. vaccae may be more effective. Further work is required to link the ability of M. vaccae to modulate protective cytokine profiles to favourable outcome in clinical studies. The high mortality found in this study suggests urgent reviews of chemotherapy and monitoring of patients are necessary in KwaZulu.