This methodologic study assessed the accuracy of family history data in ascertaining psychiatric disorders in relatives. Comparison of diagnoses based on family history with diagnoses based on direct interview indicated that the specificity for the family history method is high, but that the sensitivity is generally low. Accuracy was better for affective disorders and alcoholism than for less severe disorders; spouses and offspring provided more accurate information than parents and siblings. The use of multiple information increased sensitivity somewhat, with little adverse effect on specificity. However, because errors were often correlated when more than one person provided information about a particular relative, the use of multiple informants generally did not improve accuracy substantially. Analysis of family-genetic studies should take account of the differential quality of data obtained by the family history method vs direct interview.