The authors in this paper discuss a new approach to the assessment of physicians' performances in practice using undetected standardized (simulated) patients. Case-specific performance criteria were established for seven standardized patients by eight family physicians and two specialists. The patients were then introduced into the practices of the family physicians and of a second cohort of seven family physicians. No differences were found between the criteria-setting and the noncriteria-setting physicians; but large differences were found among the criteria, the physicians' performances as indicated by the patient, and the data recorded by the physicians on the patient chart. Depending on the method used to score the performance and the recorded data, about 30 to 45 percent of the procedures were not performed, and 50 to 70 percent of the criteria were not recorded. The implications of the study for assessment of physician performance are discussed.