Studies of alcohol-related birth defects in humans rely heavily on maternal self-reports of alcohol use. The accuracy of self-reports, and thus of related risk levels, is of concern to investigators, to public health professionals, and to persons advising pregnant women. We contrasted quantitative reports of drinking in pregnancy by 238 women with retrospective reports obtained 5 years later. Although correlated (r= 0.67, p <0.0001), marked discrepancies suggested underreporting during pregnancy by a significant proportion of the women. The validity of the retrospective index was indicated in that it was somewhat more highly related to scores on the Michigan Alcoholism Screening Test (MAST) and to a tally of alcohol-related craniofacial anomalies in the neonate than was the in-pregnancy index. Furthermore, the retrospective index and the MAST score were significant predictors of other neonatal anomaties; the in-pregnancy index was not. The only variable related to underreporting was MAST score, i.e., the higher the MAST score the greater the underreporting. The results suggest that previously reported thresholds of effect based on self-report data may be underestimates. The results also suggest that questions regarding alcohol-related problems may be more effective than direct consumption questions in the identification of women who drink heavily in pregnancy.