Assessment of alcohol-related fetal risk is complicated by difficulties in assessing alcohol exposure. We previously described considerable underreporting of alcohol use during pregnancy, particularly among problem drinkers. This report is of the relationship between severity of drinking problems and the magnitude of underreporting. Pregnant disadvantaged women responded to the Michigan Alcoholism Screening Test (MAST) at the first antenatal visit and provided self-report data regarding alcohol use at each antenatal visit. Five years later, 238 of these women again provided self-report information about drinking during the target pregnancy. Underreporting was defined as a positive difference between the indices computed from the retrospective and the in-pregnancy data. The relationship of underreporting to the MAST score was highly significant and indicated a 20-fold increase in underreporting from the lowest to highest MAST score. However, the marked individual variability seen in this relationship indicates that while the MAST may be a better indicator of risk than self-reports of drinking, it is also not sufficient for the purpose. Further work examining other measures and techniques for obtaining drinking information is indicated.