The objective of this study is to compare the ability of Medicare and cancer registry data to identify incident cancer cases and initial surgical therapy both singly and in combination. Data from the Virginia Cancer Registry (VCR) were linked to Medicare claims files (Medical Provider Analysis and Review File (MEDPAR)) for Virginia residents aged 65 years and over with breast, colorectal, lung, or prostate cancer diagnosed between 1986 and 1989. MEDPAR found 73-83% of cancer cases identified by VCR. Factors significantly associated with MEDPAR missing a case that was reported to VCR included younger age, male gender, living in an urban area, higher social class, in situ disease, and lack of cancer treatment. A total of 70-82% of cancer cases identified through Medicare claims were reported to the VCR. Older age, female gender, nonwhite race, comorbid conditions, no surgical procedures, multiple cancer admissions, and the position of the cancer diagnostic code on the MEDPAR record were factors significantly related to being missed by the VCR. The rate of capturing initial surgical therapies was similar to that of identifying cases. Combining information from VCR and MEDPAR resulted in increasing sensitivity for identifying incident cases to 92-97%. Using combined data from independent sources may improve reporting, increase the accuracy of cancer incidence estimates, and provide an opportunity to identify reasons for missing data.