The objective of this paper was to describe the distribution of tuberculosis (TB) mortality by area in the municipality of São Paulo, Brazil, from 1994 to 1998, and to evaluate its statistical association with several population characteristics. We surveyed TB deaths grouped by residential area, at the district level, and we calculated the rates for these areas standardized by gender and age groups. We applied simultaneous autoregressive--SAR regression analysis (autocorrelated errors model) in order to fit a "stepwise" model correlating TB deaths with the variables of interest. Significant associations were found between TB mortality rates and AIDS mortality rates, overcrowding at the household level, social development (expressed by a socioeconomic index), and rates of foreign immigration and immigration from other Brazilian States. Regression analysis allowed us to estimate the frequency of TB deaths virtually attributable to co-infection with HIV at 22.37% (95% confidence interval: 12.15-41.17%). TB death rates and utilization of public health services were not statistically associated, suggesting a reduced effectiveness of programs directed at control of the disease. The correlation between TB death rates and deprivation, measured by the socioeconomic index, indicates higher mortality in underprivileged areas. The significance of the association between housing overcrowding and TB deaths, in contrast to the absence of association with district-level overcrowding, indicates that prolonged contact is needed for disease transmission. Although the influx of foreigners and national migrants to the city diminished after the 1980s, immigration rates have been significantly correlated with TB mortality, suggesting greater vulnerability of these population segments to the disease.