Background: The aim of the present study was to examine the association between antidepressant use, diagnosed depression, and new-onset diabetes among elderly Medicare beneficiaries. Methods: Longitudinal data from merged survey and claims from the nationally representative Medicare Current Beneficiary Survey(MCBS) from 1999 to 2005 were used. Diabetes incidence was extracted from claims and survey data over a 3-year period. Data regarding depression and antidepressant use over time were obtained. Multivariate logistic regression analysis was used to examine associations between antidepressant use, depression, and new-onset diabetes, adjusted for demographic, socioeconomic, and lifestyle risk factors. Analyses accounted for the complex design of the MCBS. Results: The incident diabetes rate was 4.8% for those "without depression and without antidepressants" and 9.5% for those with any antidepressant use in all 3 years and diagnosed depression". Compared with Medicare beneficiaries who did not report any antidepressant use, beneficiaries reporting antidepressant use in all 3 years were 50% more likely to have new-onset diabetes. However, when diagnosed depression was entered in the model, there was no significant association between long-term antidepressant use and new-onset diabetes. Medicare beneficiaries with any depression were twice as likely as those without depression to develop diabetes (adjusted odds ratio 2.04; 95% confidence interval 1.51, 2.75). Conclusions: Depression independently increased the risk of developing diabetes in the MCBS population, although there is no evidence of an association between antidepressant use and new-onset diabetes. If replicated, these results have significant clinical implications.