BACKGROUND. The patient-physician relationship is central to medical practice. Increasingly, educators and certifying bodies seek to assess trainees' humanistic qualities. METHOD. The humanistic qualities of first-year internal medicine residents were rated in 1987–88 and 1988–89 by patients hospitalized on the general internal medicine and pulmonary services of the University of Michigan Hospital. Attending physicians (for 1988–89 only), program supervisors (program directors and chief residents), and nurses (for 1988–89 only) rated the same residents, and these ratings were compared with those of the patients. RESULTS. A total of 625 patient questionnaires for 70 residents were analyzed, with a mean of nine patient evaluations per resident and a range from four to 24. Analysis showed that more than 50 patients would need to rate each resident to achieve desired levels of reproducibility. Large numbers of attending physicians (20 to 50) would also be required to obtain a reproducible assessment; the attending physicians' ratings correlated only moderated well (r = .26) with the patients' ratings. Ratings from smaller numbers of program supervisors (five to ten) and nurses (ten to 20) would be needed for reproducible assessments. However, only the nurses' ratings showed a moderately strong relationship (r = .35) with the patients' ratings. CONCLUSIONS. Patients, attending physicians, program supervisors, and nurses view differently the humanistic attributes of residents as they interact with patients. Large numbers of patients and attending physicians would be needed to obtain reproducible ratings. Nurses' and program supervisors' ratings are much more reproducible, but nurses' perceptions correlate more closely to those of patients.