This work aimed at comparing the accuracy of the psychiatric diagnoses made under indirect supervision to the diagnoses obtained through Structured Clinical Interview for DSM-III-R (SCID). The study was conducted in 3 university services (outpatient, inpatient and emergency). Data from the emergency service were collected 3 years later, after changes in the training process of the medical staff in psychiatric diagnosis. The sensitivity for Major Depression (outpatient 10.0%; inpatients 60.0%, emergency 90.0%) and Schizophrenia (44.4%; 55.0%; 80.0%) improved over time. The reliability was poor in the outpatient service (Kw = 0.18), and at admission to the inpatient service (Kw = 0.38). The diagnosis elaborated in the discharge of the inpatient service (Kw = 0.55) and in the emergency service (Kw = 0.63) was good. Systematic training of supervisors and residents in operational diagnostic criteria increased the accuracy of psychiatric diagnoses elaborated under indirect supervision, although excellent reliability was not achieved.