BACKGROUND: Retrospective claims databases are commonly used in outcomes research. Since physician charts are rarely available to confirm diagnoses, care must be taken when choosing patient populations. OBJECTIVES: To show how patient selection criteria affects sample size and chemotherapy treatment rates using an ovarian cancer (OC) population. METHODS: Patients were initially selected if they met the following inclusion criteria: at least one diagnosis of OC (ICD-9-CM codes 183.0x) between 1/1/2002 and 12/31/2007 (first OC diagnosis date termed index), 6 months pre- and 12 months post-index eligibility, and no OC diagnosis in the 6 months pre-index. Additional criteria were imposed to further refine the sample and assess variations in chemotherapy treatment rates. First, patients were required to have at least two diagnoses of OC at least 14 days apart. Next, patients were required to have both OC diagnoses on a record labeled as medical, surgical, facility, or pharmacy (i.e., ancillary records were excluded).  RESULTS: A total of 37,172 patients had at least one diagnosis of OC. Of those, 16,418 had 6 months pre- and 12 months post-index eligibility with no pre-index OC diagnoses. In this population, 26% of patients received chemotherapy. When patients were also required to have one additional OC diagnosis at least 14 days from index, the sample size dropped to 7,431 patients, of whom 47% received chemotherapy. When OC diagnoses on ancillary records were excluded, a total of 6,213 patients were identified, of whom 52% received chemotherapy. CONCLUSIONS: Chemotherapy rates among OC patients varied significantly by the sample selection criteria used. Care must be taken to identify the correct patient sample in any retrospective database analysis since selection criteria affects the appropriateness of the sample, and thus the study results.