To determine the incidence of acute complications of an operation in a large cross-section of U.S. community hospitals, we developed a method that combines the screening of discharge abstracts with the sampling of medical records. Our target patients were those who underwent discectomy for routine lumbar discogenic radiculopathy. We assembled 39,389 discharge abstracts of patients who underwent lumbar discectomy in the U.S. in 1980. This population contained some nontarget patients with a) errors in principal diagnosis or procedure, b) operations exceeding two levels, c) fusion, d) previous laminectomy, or e) problems of the lumbar spine in excess of routine disc disease. Screening of discharge abstracts allowed us to exclude some of these patients. Those remaining were then divided by the reported hospitalization into three groups: 1) death, 2) nonfatal complication, and 3) normal hospitalization. A sample of patients from each group was selected for detailed study, and questionnaires were mailed to hospital medical records departments to confirm the hospitalization and to obtain other medical information unavailable in the With this information, we 1) removed the remaining nontarget patients a through e above); 2) categorized reported complications as erroneous, preexistent, trivial or major; 3) calculated the incidence of major complications; and 4) assessed the reliability of the discharge abstract data. Principal diagnosis and procedure were reliably coded in 96.5% of abstracts. No deaths were reported in error and none were excluded in error. Among patients who had complications, 4% of the abstracts incorrectly indicated complications, and 11% omitted one complication while correctly reporting others. Among the abstracts reporting a normal hospitalization, 0.7% omitted a complication. The abstract data had a 62% sensitivity and a 99% specificity for complications. Based on a final clean population of 28,395 patients, the incidence of death (per 10,000) was 5.9, and that of serious complication was 157.