The authors explored the factors influencing the agreement of diagnoses of alcoholism obtained by a best-estimate (BE) procedure versus those obtained by family history (FH) only, based on data from the Roscommon Family Study. The participants were first-degree relatives of either schizophrenic subjects, subjects with affective disorders, or matched community controls. The FH information was obtained from first-degree relatives of the participants, whereas the BE diagnoses included personal interview and medical records as well as FH information. Two types of error were distinguished: false-negative FHs, who were diagnosed with alcoholism by BE but not by FH; and false-positive FHs, who were diagnosed with alcoholism by FH but not by BE. The risk of false-negative FHs was increased by young age of the subject and male gender of the informant, and decreased by a history of previous hospitalization of the subject. Conversely, the risk for false-positive FHs was increased by older age of the subject, male gender of the subject, female gender of the informant, and informant's diagnosis of alcoholism. Comorbid diagnosis of nonaffective psychosis increased the risk of both types of error. It is concluded that when validated against a BE diagnosis, the FH diagnosis of alcoholism is subject to several biases and that the FH method is not a satisfactory substitute for BE diagnoses.