To explore the relationship between light to moderate alcohol consumption and risk of dementia and Alzheimer's disease in very old people, a community-based dementia-free cohort (n = 402) was followed for almost 6 years to detect incident dementia using the Diagnostic and Statistical Manual of Mental Disorders, 3rd Edition-Revised criteria. Data from the entire cohort and a subpopulation of those with baseline Mini-Mental State Examination score > or =24 (n = 317) were analyzed with Cox models. In the entire population, light to moderate drinking was significantly associated with a decreased risk of incident dementia and Alzheimer's disease compared with nondrinking (adjusted relative risk 0.5, 95% confidence interval 0.3 to 0.7). In the analysis of the subpopulation, however, the inverse association between light to moderate drinking and risk of incident dementia and Alzheimer's disease was less evident and no longer statistically significant. This study suggested that light to moderate alcohol drinking might protect against dementia and Alzheimer's disease among old people, although the possibility that such an association may be due to information bias cannot be totally ruled out.