BACKGROUND Arsenic is a potent human carcinogen and toxicant. Elevated concentration of arsenic in drinking water is a major public-health problem worldwide. We evaluated risks of adult mortality (due to cancer and cardiovascular and infectious diseases) in relation to arsenic exposure through drinking water. METHODS A cohort analysis was applied to survival data prospectively collected during 1991-2000 in a health and demographic surveillance system in Matlab, Bangladesh, where tubewells were installed beginning in the early 1970s. A total of 115,903 persons aged 15 or more years on 1 January 1991 were available for analysis. In this period, 9015 people died and 22,488 were lost to follow-up. Arsenic exposure data were derived from a survey in 2002-2003 of past and current water use and arsenic concentrations in all tubewells. We estimated risk of excess mortality in relation to arsenic exposure, using proportional hazards models. RESULTS Even at low levels (10-49 mug/L) of arsenic in drinking water, we observed increased risk of death due to all nonaccidental causes (hazard ratio = 1.16 [95% confidence interval = 1.06-1.26]). Increased risks at exposure of 50-149 microg/L were observed for death due to cancers (1.44 [1.06-1.95]), cardiovascular disease (1.16 [0.96-1.40]), and infectious diseases (1.30 [1.13-1.49]). We observed clear dose-response relationships for each of these causes. CONCLUSIONS Arsenic exposure through drinking water has generated excess adult mortality after 20-30 years of exposure.