BACKGROUND: All-cause mortality, based on national tuberculosis programme (NTP) register deaths, may under- or overestimate tuberculosis (TB) specific mortality in the population. OBJECTIVE: To assess the factors influencing this measurement in a single large population with high TB prevalence and mortality. METHODS: Routinely collected data on TB cases and treatment outcomes were linked to population data from a cohort of South African miners from 1995 to 2008. Vital status and cause of death were determined from multiple sources, including the TB programme, death register and autopsy. RESULTS: The TB mortality rate, based on 430 deaths on the TB register, was 192/100,000 person-years (py). Many of these deaths (57%) were not caused by TB, and 483 TB deaths were identified outside the programme. Overall, there were 674 TB-specific deaths; the TB-specific mortality rate was 302/100,000 py. These deaths included 191 (28%) on the TB register, 23 (3%) among defaulters/transfers, 153 (23%) after anti-tuberculosis treatment and 307 (46%) in men who had never been on the programme. CONCLUSIONS: This study highlights methodological issues in estimating TB mortality. In this population, a method using the product of TB incidence and case fatality consistently underestimated TB mortality. Accurate estimates of TB-specific mortality are crucial for the proper evaluation of TB control programmes.