Alcoholism is a possible cause of dementia, mainly through associated nutritional deficiencies and, rarely, through acute direct toxicity. However alcohol consumption was not found to be a risk factor in previous epidemiologic studies. We prospectively studied 3,777 community residents aged 65 and over, in the districts of Gironde and Dordogne. Average daily alcoholic consumption was recorded at baseline. Incident cases of dementia and Alzheimer's disease were screened at follow-up with explicit criteria. At 3 years, 2,273 subjects not demented at baseline were still available for follow-up. Wine was the only alcoholic beverage reported by more than 95 p. 100 of regular drinkers. In the 318 subjects drinking 3 to 4 standard glasses per day (> 250 and up to 500 ml), categorized as moderate drinkers, the crude odds ratio (OR) was 0.18 for incident dementia (p < 0.01) and 0.25 for Alzheimer's disease (p < 0.03), as compared to the 971 non-drinkers. After adjusting for age, sex, education, occupation, baseline MMSE and other possible confounders, the ORs were respectively 0.19 (p < 0.01) and 0.28 (p < 0.05). In the 922 mild drinkers (< 1 to 2 glasses per day) there was a negative association only with AD, after adjustment (OR = 0.55; p < 0.05). The inverse relationship between moderate wine drinking and incident dementia was explained neither by known predictors of dementia nor by medical, psychological or socio-familial factors. Considering also the well documented negative associations between moderate wine consumption and cardiovascular morbidity and mortality in this age group, it seems that there is no medical rationale to advise people over 65 to quit drinking wine moderately, as this habit carries no specific risk and may even be of some benefit for their health. Advising all elderly people to drink wine regularly for prevention of dementia would be however premature at this stage.