In their survey of prescription medication sharing, Goldsworthy et al.1 appropriately noted the paucity of information on this important behavior. Given concerns about potentially harmful effects on the fetus from in utero exposures to medication, it is particularly important to understand medication-sharing behaviors among a broad sample of pregnant women. We examined this question some time ago. A performer participates in the second Walk Against Children's Malnutrition and Hunger in El Salvador, organized by the World Food Programme of the United Nations in the face of the global food crisis. Photograph by Jose Cabezas/AFP/Getty Images. Printed with permission. In the early days of our current case–control surveillance study of birth defects in relation to medication use in pregnancy,2 we evaluated the effect of questionnaire design on recall of medication use. As part of that analysis, we summarized data from interviews conducted between 1976 and 1984 involving 5435 mothers who delivered malformed or nonmalformed infants in geographic areas surrounding Boston, Massachusetts; Philadelphia, Pennsylvania; Toronto, Ontario; and in part of Iowa. Interviews were conducted within 6 months of delivery, and included questions about use in pregnancy of 9 specifically named prescription medications. Women who answered affirmatively to any of those drugs were asked whether these medications were obtained from physicians or from other sources, such as friends, neighbors, or relatives.3 Among users of Bendectin, Fiorinal, Hydrodiuril, Clomid, and Seconal, less than 5% reported a source other than a physician; however, higher proportions reported nonphysician sources for Compazine (9%), Diuril (11%), Valium (18%), and Darvon (22%). Goldsworthy et al.1 correctly noted a number of consequences of medication-sharing behaviors, to which we would add an additional consequence: pharmacoepidemiology studies of drug safety increasingly rely on exposure information drawn from computerized prescription or pharmacy records (often as part of HMO or claims data). Our earlier conclusion, now supported by the work of Goldsworthy et al., is therefore even more relevant today: [I]n addition to concern that [information from such sources] includes false positive ‘exposures’—prescriptions issued or filled but not consumed—our findings suggest that these data sources may also include appreciable numbers of false negative exposures—drugs consumed but obtained from nonidentified sources.3(p675) Appreciable exposure misclassification would have very serious consequences when it comes to using prescription data sources to identify adverse effects of medication exposures; any such efforts must take into consideration the extent to which studied medications are shared.