BACKGROUND : A retrospective cohort study was carried out to compare the survival between smear-positive patients and smear-negative multidrug resistant tuberculosis (MDR-TB) patients hospitalised in a specialised TB hospital in Witbank, South Africa. METHODS : A review of medical records of MDR-TB patients treated from 2001 to 2010 was carried out. Survival time was measured from a patient's date of hospitalisation to the date when the patient died, was last treated at the hospital or the end of the study (whichever came first). All patients who were alive until the end of the study period or lost to follow-up were censored and those who died were considered as failures. Survival patterns were estimated using Kaplan Meier plots, log rank tests and life tables. Cox proportional hazards regression analyses were also conducted. RESULTS : The mean age of the 442 MDR-TB patients in the study was 37.7 ± 11.2 years. The incidence rates of mortality were 13.4 and 43.9 per 1000 person-months for smear-negative and smear-positive MDR-TB patients, respectively. Cox proportional hazard regression showed that the predictors of death among MDR-TB patients include HIV co-infection (adjusted Hazard Rate, aHR = 1.89, 95% CI = 1.02-3.52), old age (above 60 years) (aHR = 2.05, 95% CI = 1.04-3.60) and smear positivity at diagnosis (aHR = 3.29, 95% CI = 2.39-4.64). CONCLUSION : The study showed that the probability of survival during the treatment is reduced in MDR-TB patients, who are smear-positive, HIV positive or older than 60 years. Special care should be given to these patients to improve survival.