Global ratings of dementia severity are used increasingly in clinical trials of antidementia compounds. Such ratings are clinically relevant, but their reliability in multicenter settings has not been determined. To evaluate the reliability of one global scale, the Clinical Dementia Rating (CDR), 82 investigators of the multicenter Alzheimer's Disease Cooperative Study participated in a training and reliability protocol using videotaped assessments of subjects in various stages of Alzheimer's disease. Following training, overall agreement of the investigators with 'gold standard' CDR scores was 83%. These results indicate that the training protocol is useful for establishing good levels of agreement in staging dementia severity and that the CDR can be standardized as a clinical global scale for multicenter studies of Alzheimer's disease.