Objective. - To report on data collected during on-site audits of source documents in the Cancer and Leukemia Group B (CALGB). Design. - A retrospective review of audit reports in four audit cycles. Setting. - A cooperative group of institutions conducting clinical trials in cancer treatment. Participants. - Patients taking part in clinical trials at collaborating CALGB institutions, members of the CALGB Data Audit Committee, and group chairmen of CALGB. Main Outcome Measure. - The results of 691 institutional audits conducted by the CALGB in 1982 through 1992 with comparisons of main CALGB institutions vs affiliates. Results. - In four full reviews of all participating institutions in the CALGB, 3787 patients have had their on-site medical records compared with data submitted to the CALGB Data Management Center. Compliance with federal regulations for oversight by an institutional review board improved from a deficiency rate of 28.0% among the main institutions and 49.6% of the affiliate institutions in the first audit cycle to respective figures of 13.3% and 28.2% in the fourth cycle. Consent form deficiencies also dropped overall from 18.5% in the first cycle to 3.9% in the fourth. Patient eligibility was verified by auditors in 94.5%, and assessment of tumor changes in response to treatment was verified in 96.4% in the fourth cycle; both figures were only slightly lower in the first cycle. Two instances of scientific impropriety were discovered for a rate of only 0.28% of all audits. Both occurred prior to 1984, and none have occurred since. Major protocol deviations in drug dosing have held steady at about 11% over four audit cycles. Over the 11-year period of audits, three main institutions and 96 affiliate institutions have discontinued CALGB membership due solely, or at least partly, to unfavorable audit results. Conclusion. - Scientific improprieties have occurred very rarely in clinical trials conducted by the CALGB. Protocol compliance in assessing patient eligibility and tumor responses has been high. Attention to administrative matters of consent forms, institutional review board approval, and ancillary data submission has measurably improved in the CALGB, which is at least partly due to the pressure from this on-site peer review of investigator performance.