Abstract Introduction Osteonecrosis (ON) is a rare disease associated with alcohol and glucocorticoid use. Identifying additional risk factors is difficult as the number of cases at any single center is small. We investigated whether data available in large health care databases can be used to identify incident ON cases. Methods Using data from the Boston Veterans Affairs Healthcare system, we identified potential cases of ON. These records, including available radiographs and reports, were reviewed. Using published criteria, we evaluated whether the subjects had confirmed ON (radiographs/reports met criteria), incident ON (onset of symptoms within 6 months of first code), or prevalent ON (onset more than 6 months prior to first code or onset could not be determined). We tested different definitions for incident ON using information derived from administrative data. These were compared to the 'gold standard' (record review) and positive predictive values (PPVs) were derived. Since PPVs for incident cases were low, we found the number of incident cases expected for every 1,000 potential cases identified, using the definitions as an initial screening tool to reduce the number of medical records that required examination. Results We identified 87 potential cases. No case of jaw ON was identified. Only 15 (17%) incident cases of ON were identified. PPVs never exceed 50% for incident ON. However, if we used the definition '(at least 1 inpatient ON code) and (no prior codes for osteoarthritis)' as an initial screen, then for every 1,000 records, we would need to review only 150 to find 69 incident cases. Conclusions Though the precise PPVs we found may not be generalizable to other databases, we believe that administrative data alone should not be used to identify incident cases of ON without confirming the diagnosis through a review of medical records. By applying the above definition, the number of records requiring review can be markedly reduced. This method can be used to find cases for valid case-control studies of risk factors for ON.