Administrative data have long been used in psychiatric epidemiology and outcomes evaluation. This article examines the reliability of the recording of schizophrenia and depressive disorder in three Saskatchewan administrative health care utilization datafiles. Due to their comprehensive nature, these datafiles have been used in a wide range of epidemiologic studies. Close agreement was found between hospital computer data and patients' charts for personal and demographic factors (> or = 94.7%). Diagnostic concordance between computerized hospital data and medical charts was very good for schizophrenia (94%) but poor for depressive disorder (58%). Appropriate physician services were identified for 60% and 72% of hospital discharges for schizophrenia and depressive disorder, respectively, and exact diagnostic agreement between hospital and physician datafiles was 62% for schizophrenia and 66% for depressive disorder. Appropriate provincial mental health branch services were found for 83% and 38% of hospital discharges for schizophrenia and depressive disorder, respectively; exact diagnostic concordance between these datafiles was 75% for schizophrenia and 0% for depressive disorder. A significant number of patients with major or neurotic depression appeared to be wrongly coded as having depressive disorder in the hospital file. The differences in diagnostic agreement may also be partly a function of how the two conditions are differentially treated in the health system. These findings suggest that more specific and severe psychiatric diagnoses are likely to be recorded accurately and consistently in the Saskatchewan datafiles. However, disorders with multiple manifestations or those for which there are several possible codes should be examined with caution and ways sought to validate them. Attention should also be paid to which service sectors are involved in the treatment of specific disorders.