Epidemiologic study of Alzheimer's disease by family history requires that Alzheimer's be distinguished from other dementias. Identification of demented family members is usually based on recall by relatives. This study examines the validity of the classification of Alzheimer's disease and other dementias based on relatives' reports. Close relatives of autopsy-confirmed dementia cases were asked to complete a questionnaire describing the patient's symptoms. These informants were familiar with the patient's disease and involved in his/her care prior to death. The questionnaire included the DSM-III criteria for Primary Degenerative Dementia and the Hachinski Ischemic Scale. A diagnosis derived from the close relatives' responses was compared to the neuropathologic diagnosis for thirty-six cases: 20 Alzheimer's disease, 9 mixed Alzheimer's disease, and 7 non-Alzheimer's disease dementias. The diagnosis of Primary Degenerative Dementia derived from questionnaire responses had a sensitivity of 0.93 and specificity of 0.43 for pathologic Alzheimer's disease. Few vascular dementias were included in the series, thereby precluding the study of so-called multi-infarct dementia. Hachinski scores based on relatives' responses classified 40% of pathologically pure Alzheimer's disease cases as multi-infarct dementia (HIS greater than 7). Thus, using these elevated Hachinski scores to rule out Alzheimer's disease would cause substantial misclassification. Diagnosis based on questionnaire Primary Degenerative Dementia criteria was quite sensitive but relatively nonspecific. When attempting to obtain a complete family history or pedigree that describes the occurrence of Alzheimer's disease and other dementias in all family members, the questionnaire approach should be supplemented with additional information from medical records, physicians, and other relatives.