Research on the relation between diabetes mellitus and dementia has produced conflicting results, and the relation has not been investigated among Blacks and Hispanics. In this study, Cox proportional hazards models were used to analyze longitudinal data from 1,262 elderly subjects without dementia at baseline (1991-1996) who were followed for an average of 4.3 years between 1992 and 1997. Outcomes were incident Alzheimer's disease and dementia associated with stroke. The prevalence of diabetes was 20% at baseline. The adjusted relative risk of Alzheimer's disease among persons with diabetes as compared with those without diabetes was 1.3 (95% confidence interval (CI): 0.8, 1.9). The adjusted relative risk for the composite outcome of Alzheimer's disease and cognitive impairment without dementia (without stroke) in subjects with diabetes was 1.6 (95% CI: 1.2, 2.1). The adjusted relative risk of stroke-associated dementia in persons with diabetes was 3.4 (95% CI: 1.7, 6.9). Among Blacks and Hispanics, approximately one third of the risk of stroke-associated dementia was attributable to diabetes (33% (95% CI: 31, 36) and 36% (95% CI: 33, 37), respectively), as compared with 17% (95% CI: 13, 22) among Whites. The finding of an association between diabetes and the composite outcome of Alzheimer's disease and cognitive impairment without dementia (without stroke) is consistent with prior reports of a modest relation between diabetes and Alzheimer's disease.