Although the NINCDS-ADRDA Work Group has recently developed uniform clinical criteria for the diagnosis of Alzheimer's disease (AD), inter-rater reliability studies of these criteria are few. We report a study in which 2 neurologists and 2 psychiatrists independently reviewed clinical data abstracted from the records of 30 demented subjects and 10 nondemented control subjects participating in a longitudinal study of AD at the University of Pittsburgh. We recorded the clinical data on a standardized form; the subjects' identity and clinical and pathologic diagnoses were omitted. Each physician diagnosed each case according to the NINCDS-ADRDA criteria. We calculated the inter-rater agreement for all possible 2-way combinations of clinicians with the Kappa statistic, which ranged from 0.36 (fair agreement) to 0.65 (substantial agreement). We conclude that current NINCDS-ADRDA criteria enable moderate levels of agreement among clinicians in general.