The authors, as part of the ongoing NIMH Collaborative Study on the Psychobiology of Depression, used an analysis of variance design and videotaped interviews to explore the effects of sources of variance on the reliability of the measures being used by the NIMH study. In spite of substantial differences among interviewers in background or orientation, the authors found that diagnoses and symptom ratings were made with a high level of reliability. These results suggest that the use of structured interviews and diagnostic criteria, when combined with a careful and systematic training program, can lead to good levels of diagnostic reliability.