We linked Medicare claims data to information on 417 patients with a clinical diagnosis of Alzheimer's disease in the Consortium to Establish a Registry for Alzheimer's Disease (CERAD) to determine what proportion of them were identified as having Alzheimer's disease (AD) in Medicare claims records. Seventy-nine percent of these patients were identified as having AD using 5 years of claims data; 87% were identified as demented when a broader set of ICD-9-CM codes was used. An Anderson-Gill counting process approach was used to model the "hazard" of patients being identified as having AD in Medicare claims data. CERAD patients with mild dementia were less likely to be identified in the claims data as having AD. Once identified in Medicare claims as having AD, patients were more likely to be so identified again. When using only the physician supplier and institutional outpatient files, approximately 75% of CERAD patients were identified as having AD; hospital files used alone identified less than one-third (29%) of the CERAD patients as having AD. The data indicate that at least 3 consecutive years of physician supplier and physician outpatient claim files should be used to identify Medicare beneficiaries with AD using Medicare claims.