BACKGROUND: The accuracy and completeness of the SEER-Medicare data for measuring cancer-related therapy have not been extensively evaluated. OBJECTIVES: To investigate the best method for measuring cancer-related surgery among patients in the SEER-Medicare database. SUBJECTS: A total of 149,970 incident cases of breast, colorectal, endometrial, lung, pancreatic, and prostate cancer diagnosed between 1991 and 1993.MEASURES The most invasive surgical procedure identified through Medicare's inpatient, physician, and hospital outpatient claims was compared with corresponding data from the SEER files. RESULTS: Agreement between the SEER and Medicare files was generally highest for resection and radical surgery (eg, kappa 0.70-0.90). While there was less agreement regarding no surgical therapy and biopsy individually, the concordance of the two sources in excluding cancer-directed surgery was high. Compared with inpatient data alone, using the combined inpatient, physician, and outpatient data increased concordance between SEER and Medicare for less invasive procedures. CONCLUSIONS: The agreement of SEER and Medicare data appears to be good for major surgical procedures and for excluding persons who did not undergo cancer-directed surgery. Both the SEER and the Medicare data captured a small number of surgeries not reported in the other file. Therefore, where possible, using both data sources will enhance identification of surgeries. Because the analysis was performed with linked data, the accuracy of surgical claims in Medicare data alone cannot be assessed.