There are standardized criteria to assist in the diagnosis of Alzheimer's disease (AD), a disorder that lacks unique clinical, morphologic, or biochemical features. Diagnostic reliability of single groups of investigators using these criteria is moderate to substantial. In this study, seven clinicians at separate sites established a criteria-based diagnosis in 42 consecutive memory disorder patients participating in a national genetic epidemiologic study using a quantitative multiaxis AD rating scale (ADRS) that incorporates NINCDS/ADRDA criteria, reliability of information, and comorbidity. Reliability, measured by a generalized kappa statistic for more than two raters, was substantial (0.63 +/- 0.13) when the subjects were grouped as "AD" (probable or possible) versus "not AD," but somewhat lower (0.52 +/- 0.10) when subjects were classified as probable AD, possible AD, or not AD. There was unanimous agreement for two-thirds of the subjects using a dichotomous classification scheme. These findings suggest that the ADRS is a useful diagnostic instrument for multicenter studies.