OBJECTIVES: In retrospective claims database studies, defining the sample often receives less attention than the statistical methods used in the analysis. However, identifying an appropriate sample is a key step in the analytic process. The current research was undertaken to identify typical sample selection definitions in retrospective claims database studies of mood disorders. METHODS: A PubMed literature search was conducted to identify January 2000 to September 2005 publications with the major topic heading of mood disorders and terms such as administrative data, insurance claims or claims analysis. The criteria used for sample selection was abstracted from relevant studies. RESULTS: Forty-seven studies were abstracted, including 10 specifically focusing on major depressive disorder (MDD), 15 on depression and 13 on bipolar disorder (BPD). No set of ICD-9-CM diagnosis codes was used to identify these conditions across studies. Three studies defined MDD as 296.2 or 293.3 only, while others included codes for dysthymia (300.4) and depressive disorder not otherwise specified (311). Some used codes for adjustment reaction with depressive symptoms (309.0, 309.2). One study of depression used only 296.2 and 296.3, while another used nine different 4-digit ICD-9 codes. All BPD studies that specified codes used 296.4 - 296.6, but the inclusion of other 296 codes varied by study. Several publications did not list the specific diagnosis codes used. Some studies required at least two diagnoses, while others required only one. About half the studies provided little basis for the coding criteria used. CONCLUSIONS: A specific operational definition of mood disorders was not evident from the literature. Results of retrospective claims database studies with different sample selection criteria may vary, so it is important to assess findings in light of the sample definition.