One hundred twenty patients admitted to the medical service and presented at one of two teaching conferences were examined to evaluate the nature of errors made in physical examination by house officers. The overall mean error rate per patient was 5.8% +/- 0.6% for residents and 6.4% +/- 0.6% for interns. The mean error rate for cardiac examination was significantly worse than for respiratory, abdominal, or thyroid examinations. An additional 40 patients selected for abnormal cardiac or neurologic findings were found to have a significantly higher mean error rate in those areas than did the first group of patients with less abnormal findings. Agreement in physical findings between investigators was significantly better than between investigators and house officers. Resident performance in physical examination correlated with estimates of relative time spent by attending physicians at the bedside reevaluating physical findings, but not with class-ranking and inservice examination scores. Educational implications are discussed.