Summary Dementia and non-insulin-dependent diabetes mellitus (NIDDM) are highly prevalent disorders in the elderly. Diabetes has repeatedly been reported to affect cognition, but its relation with dementia is uncertain. We therefore studied the association between diabetes and dementia in the Rotterdam Study, a large population-based study in the elderly. Of 6330 participants, aged 55 to 99 years old, complete information on diabetes and presence of dementia was available. Diabetes was diagnosed as use of anti-diabetes medication or random or post-load serum glucose over 11 mmol/l. Dementia was diagnosed through a stepped approach, including a sensitive screening of all participants and a comprehensive diagnostic work-up. Diabetes was present in 724 (11.4 %) subjects. Of the 265 dementia patients 59 (22.3 %) had diabetes. Multiple logistic regression analyses, adjusting for age and sex differences, revealed a positive association between diabetes and dementia (odds ratio: 1.3, 95 % confidence interval: 1.0–1.9). In particular, strong associations were found between dementia and diabetes treated with insulin (odds ratio: 3.2, 95 % confidence interval: 1.4–7.5). The relation was strongest with vascular dementia, but was also observed with Alzheimer's disease. These associations were independent of educational attainment, smoking, body mass index, atherosclerosis, blood pressure and antihypertensive drug treatment, and could not be explained by clinical cerebral infarctions. The results suggest that NIDDM is associated with dementia. Alzheimer's disease may be more frequent in elderly diabetic patients treated with insulin. [Diabetologia (1996) 39: 1392–1397]