Longitudinal data collected from the Demographic Surveillance System (DSS) in Matlab, a rural area in Bangladesh, are used for determining trends in and determinants of mortality of the elderly population (60 yr and over) in 1974-1996. The old-age mortality rate is high in Matlab, 1.2 times that of Sri Lankan and 1.5 times that of the Swedish elderly population in a comparable period. Mortality among the elderly population declined in 1974 1982, but much less so in 1982-1996. Proportional hazards models were used for examining determinants of mortality in a sample of about 10,000 elderly persons. This multivariate analysis used information on several social and economic variables derived from the 1982 census and mortality data of this population which was followed prospectively in 1982-1992. Marital status was the single most important determinant: widows and widowers had 1.5 to 2 times higher risk of death compared to couples where both husbands and wives were alive. Social support in old age by children also plays a role, especially for women: women living with at least one son or daughter had 18% lower mortality than women living in a household without sons or daughters. Socioeconomic factors are also important. Those who had at least some education or were relatively affluent had lower mortality than those with no education or who were less affluent.