To verify the hypothesis that borderline diabetes may increase the risk of dementia and Alzheimer's disease, a community-based cohort of 1,173 dementia- and diabetes-free individuals aged >or=75 years was longitudinally examined three times to detect patients with dementia and Alzheimer's disease (Diagnostic and Statistical Manual of Mental Disorders, Revised Third Edition criteria). Borderline diabetes was defined as a random plasma glucose level of 7.8-11.0 mmol/l. Data were analyzed using Cox proportional hazards models. During the 9-year follow-up, 397 subjects developed dementia, including 307 Alzheimer's cases. At baseline, 47 subjects were identified with borderline diabetes. Borderline diabetes was associated with adjusted hazard ratios (95% CIs) of 1.67 (1.04-2.67) for dementia and 1.77 (1.06-2.97) for Alzheimer's disease; the significant associations were present after additional adjustment for future development of diabetes. Stratified analysis suggested a significant association between borderline diabetes and Alzheimer's disease only among noncarriers of APOE epsilon4 allele. There was an interaction between borderline diabetes and severe systolic hypertension on the risk of Alzheimer's disease (P = 0.04). We conclude that borderline diabetes is associated with increased risks of dementia and Alzheimer's disease; the risk effect is independent of the future development of diabetes. Borderline diabetes may interact with severe systolic hypertension to multiply the risk of Alzheimer's disease.