Using two population-based cohorts of men aged 45-59, we sought to derive and validate a prediction rule for identifying heavy consumers of alcohol. Eighty-five percent of eligible men on electoral rolls in Caerphilly, Wales (derivation set, N = 2512) and 90% of eligible men on the practice lists of 16 Speedwell, England, general practitioners participated (validation set, N = 2348). Alcohol consumption was assessed by questionnaire with heavy alcohol consumption defined as the top 10% of the Caerphilly population's alcohol usage (greater than 525 cc ethanol per week). The prediction rule, Score = (mean corpuscular volume x 1.00) + (body mass index x 0.31) + (systolic blood pressure x 0.08) + HDL-cholesterol x 9.24) + (fasting triglyceride x 2.20) was derived by multiple linear regression in the Caerphilly cohort and validated in the Speedwell cohort. Comparing the lower 20% of the Score distribution with the upper 5%, likelihood ratios increased from 0.15 to 5.29 and 0.06 to 7.42 in the Caerphilly and Speedwell cohorts, respectively. Having a score of 136.30 or greater yielded a relative risk of being a heavy drinker of 23.1 (95% Cl = 10.1-53.0) in Caerphilly and 99.3 (95% Cl = 12.8-769.5) in Speedwell. The derived prediction rule is a valid diagnostic aid to help clinicians identify heavy alcohol consumers.