We used cognitive deficits detected by neuropsychological testing to evaluate clinical diagnosis of Alzheimer's disease. Deficits were defined with respect to performance of control subjects according to procedural guidelines set by a NINCDS-ADRDA Work Group. The most frequent deficits were in recent memory and lexical-semantic language abilities. Clinical diagnosis of Alzheimer's disease was compared with diagnosis based on a criterion of two or more cognitive deficits both on initial neuropsychological testing and on testing repeated a year later in some subjects. Initial clinical diagnosis identified 96% of cases who met the criterion when first tested and 100% of those with multiple deficits at follow-up. Specificity with respect to the criterion was 86% on initial testing and 89% at follow-up. These findings support the validity of clinical diagnosis of Alzheimer's disease using the NINCDS-ADRDA criteria.