OBJECTIVE: Because antidepressants are prescribed for numerous conditions, using administrative databases for conducting research on depression must be done with caution. We examined the Saskatchewan Health databases, containing 3-digit ICD9 codes, concurrent medications, prescriber information, comorbidities, and fee-for service codes, to determine their suitability for conducting depression research. METHODS: We identified a cohort of individuals who began a “new” period of antidepressant use (no use 180 days prior) with at least one additional dispensation of the antidepressant. A stratified random sample of 600 individuals was chosen for medical record abstraction. The medical record abstraction classification of depression was the gold standard against which our definitions of depression were judged. We developed a primary definition of depression based on ICD9 codes and tried to further refine it with various algorithms. Since so few patients were false-positive and/or false-negative in the primary classification, it was not feasible to split the data and use pattern recognition techniques. Instead, we developed medically probable scenarios using the explanatory variables and evaluated them for their sensitivity, specificity, and predictive values. RESULTS: Defining depression with ICD9 codes 296 (affective disorders) 309 (adjustment reaction) and 311 (depressive disorders) had a sensitivity - 71%, specificity - 85%, positive predictive value (PV+) – 86%, and negative predictive value (PV-) – 70%. Using other definitions: to limit the number of false-negatives raised the sensitivity to 84% and the PV- to 77%; limiting false-positives raised the specificity 90% and PV+ to 86%. The false-negatives had very similar characteristics to the true-positives. CONCLUSION: The ICD9 definition will permit future studies of depression to be conducted using the Saskatchewan Health databases.