AIMS To test the hypothesis that the association previously reported between moderate alcohol use and better cognition is an artefact of confounding by (a) higher premorbid education and socioeconomic status; (b) a lifestyle of moderation (using smoking as a risk marker); and (c) decreased alcohol consumption in people with physical illnesses. METHOD Data were analysed from people aged 60-74 years interviewed for the 2000 British National Psychiatric Morbidity Survey, representative of people living in private homes. Alcohol use information was available for 1985 (98.9%) of the eligible participants, of whom 1735 (87.4%) who drank moderately or abstained were included in the analyses. Our main outcome measures were the Alcohol Use Disorders Identification Test (AUDIT), the Telephone Interview for Cognitive Status Screen for Cognitive Impairment and the National Adult Reading Test to measure crystallised (premorbid) intelligence. Our physical health measures were the number of prescribed medications and physical illness reported, and the 12 item Short Form Health Survey's Physical Component Score. RESULTS The relationship between current cognition and alcohol use was reduced and no longer significant after considering premorbid intelligence or physical health. In our final model, the significant predictors of current cognition among non-problem drinkers were: age (B = -0.13, -0.18 to -0.08; p<0.001) and crystallised intelligence (B = 0.14, 0.12 to 0.17; p<0.001). Smoking was not associated with cognition. CONCLUSIONS In people who were not problem drinkers, higher alcohol intake was not associated with improved current cognition after controlling for premorbid intelligence and physical health. Our findings suggest that, despite previous suggestions, moderate alcohol consumption does not protect older people from cognitive decline.