Reliability of alcohol consumption reported on a self-administered questionnaire was examined in 73 gynecologic outpatients. The questionnaire included quantity-frequency questions on current beverage-specific alcohol use (wine, beer and liquor). The questions were replicated 5 years later for both the earlier period and for current drinking patterns. Results indicated that, in general, original and retrospective data were highly correlated. Only recall of wine, the least frequently drunk beverage, was more highly correlated with current than with original consumption. Frequency of drinking was typically recalled better than quantity per occasion. Beverage-specific absolute alcohol per day (AA) was less reliable than total AA. Retrospective AA values accounted for 86% of the variability in original scores. The heaviest drinkers, however, tended to report disproportionately greater alcohol intake in retrospect. Thus, almost twice as many heavy drinkers were identified on the basis of their retrospective reports compared to those identified by their original reports. Results from this study suggest that considerable confidence can be placed in retrospective reports of total alcohol consumption by nonalcoholic women over a relatively long-term interval and that heavy drinking may be reported more accurately retrospectively than currently.