Standardization of diagnostic procedures for cognitive impairment in large epidemiologic surveys remains difficult. This paper reports results of diagnostic standardization in a subsample of 2,914 elderly (age 65 years+) Canadians from the Canadian Study of Health and Aging (CSHA; n = 10,263). The objectives were to measure the consistency of the CSHA diagnosis as a test of validity; to assess inter-rater reliability, and to assess the impact of neuropsychological data on the diagnosis of dementia. The CSHA clinical assessment included a nurse's examination, Modified Mini-Mental Status (3MS) exam and Cambridge Mental Disorders Examination, neuropsychological tests, medical history and examination, and laboratory investigations. A final diagnosis was reached in a consensus conference which incorporated preliminary diagnoses from both physicians and neuropsychologists. Computer algorithms, which were developed to check consistency between the clinical observations and the final diagnosis, demonstrated 98% concordance with DSM-III-R criteria for dementia and 92% with NINCDS-ADRADA (National Institute of Neurological and Communicative Disorders and Stroke and the Alzheimer's Disease and Related Disorders Association) criteria for probable Alzheimer's disease. Inter-rater agreement was high: kappa = 0.81 for dementia/no dementia; kappa = 0.74 for normal/cognitive impairment, not dementia/ dementia. Comparisons of diagnoses between raters by clinical specialty revealed few systematic differences. The impact of neuropsychological input on the physician's diagnosis was most marked in the borderline cases between diagnostic categories.