PURPOSE: Observational studies in Western settings show moderate alcohol use associated with better cognitive function, but they are vulnerable to contextual bias. Evidence from non-Western settings may be useful to verify causality. We examined such association in southern China where alcohol use is low. METHODS: We used multivariable linear regression in cross-sectional data from the Guangzhou Biobank Cohort Study to assess sex-stratified associations of alcohol use (never, occasional, moderate, heavy and former drinker) with delayed 10-word recall score for all 3 phases (N = 28,537) and Mini-Mental State Examination (MMSE) score in phase 3 (N = 9,571). RESULTS: Delayed 10-word recall scores were higher in moderate drinkers compared with never drinkers among men (0.30 words, 95% confidence interval [Cl]: 0.18 to 0.42) but not women (0.02; 95% Cl: -0.12 to 0.17), adjusted for sociodemographic factors. Occasional alcohol users also had higher 10-word recall scores among men (0.27; 95% Cl: 0.18 to 0.37) and women (0.30; 95% Cl: 0.23 to 0.37). These estimates were little altered by further adjustment for cardiovascular risk factors. Results for MMSE scores were similar. CONCLUSIONS: Alcohol may not drive the association between moderate use and better cognitive function, which instead may be due to confounding by general moderation in lifestyle. Ann Epidemiol 2010;20:873-882. (C) 2010 Elsevier Inc. All rights reserved.