The objective of this study was to assess whether arsenic exposure is a risk factor for diabetes mellitus as indicated in a few earlier studies. Arsenic in drinking water is known to occur in western Bangladesh, and in 1996, two of the authors conducted a survey of the prevalence of diabetes mellitus among 163 subjects with keratosis taken as exposed to arsenic and 854 unexposed individuals. Diabetes mellitus was determined by history of symptoms, previously diagnosed diabetes, glucosuria, and blood sugar level after glucose intake. The crude prevalence ratio for diabetes mellitus among keratotic subjects exposed to arsenic was 4.4 (95% confidence interval 2.5–7.7) and increased to 5.2 (95% confidence interval 2.5–10.5) after adjustment for age, sex, and body mass index. On the basis of a few earlier measurements of arsenic concentrations in drinking water by the authorities in Bangladesh and another 20 new ad hoc analyses, approximate time-weighted exposure levels to arsenic in drinking water could be estimated for each subject. Three time-weighted average exposure categories were created, i.e., less than 0.5, 0.5–1.0, and more than 1.0 mg/liter. For the unexposed subjects, the corresponding prevalence ratios were 1.0, 2.6, 3.9, and 8.8, representing a significant trend in risk (p < 0.001). The result corroborates earlier studies and suggests that arsenic exposure is a risk factor for diabetes mellitus. Am J Epidemiol 1998;148:198–203.