This study was conducted to evaluate the validity of using the Saskatchewan Health administrative claims databases for conducting depression research. To develop a claims-based definition of depression, we identified a cohort of individuals who began a "new" period of antidepressant use (no use 180 days prior) from which we selected a stratified random sample (n = 600) for medical record abstraction. The medical record diagnosis was used as the gold standard for judging our database definitions of depression. After defining a primary database definition of depression, we tried to refine it using medically probable scenarios and assessed refinement by agreement statistics. Defining depression with ICD9 codes 296 (affective disorders), 309 (adjustment reaction), and 311 (depressive disorders), the sensitivity (Se), specificity (Sp), positive (PV+) and negative predictive (PV-) values were: 71%, 85%, 86%, and 70%, respectively. Algorithms that limited the number of false-negatives resulted in: Se = 84% and PV- = 77% whereas those that limited false-positives resulted in: Sp = 90% and PV+ = 86%. Although our depression definition requires treatment with antidepressants, this definition will allow us to conduct future studies of depression and its treatment using the Saskatchewan Health databases.