Questionnaire data on alcohol consumption obtained from primary respondents participating in a community-based case-control study of coronary heart disease were compared with similar data obtained from their next of kin. The primary respondents were men and women aged 25-64 years who were a representative sample of myocardial infarction cases (n = 58) and controls (n = 456) from the case-control study. The study was conducted in Auckland, New Zealand between 1986 and 1988. When alcohol drinking frequency was collapsed into five categories, exact agreement between primary respondents and next of kin was 64% for cases and 62% for controls. Agreement within one category (plus or minus) was 89% for both cases and controls. Tests of marginal homogeneity showed no evidence that next of kin systematically over- or underestimated the primary respondents' self-reported drinking frequency. With respect to the amount of alcohol drunk, mean differences between primary respondents and next of kin were small (mean difference: 2.25 g per day for cases; -0.77 g per day for controls), although there was considerable variability between individual pairs in the level of agreement. The findings suggest that studies which use proxy sources of data to assess alcohol drinking are unlikely to produce biased estimates of alcohol consumption at the aggregate level, particularly with regard to frequency measures.