The purpose of this study is to determine sociodemographic, clinical, and pharmacotherapeutic characteristics, especially use of atypical antipsychotics, associated with incident diabetes mellitus in a population of privately insured patients with mental health diagnoses. Patients with a mental health diagnosis stably medicated for a 3-month period during January 1999 through October 2000 and having no diabetes were followed through December 2000. Cox proportional hazards models were developed to identify antipsychotic medications associated with newly diagnosed diabetes. Of the 7381 patients identified, 339 developed diabetes, representing an annual incidence rate of 4.7%. Diabetes risk among the entire sample was lowest for risperidone (hazard ratio [HR] = 0.69; p < 0.05), while quetiapine (HR = 0.74), olanzapine (HR = 0.95), and clozapine (HR = 1.22) were not significantly different from first-generation antipsychotics. Diabetes risk was significantly lower among males receiving risperidone (HR = 0.49; p < 0.01) or quetiapine (HR = 0.50; p < 0.10), while diabetes risk among females did not differ significantly from first-generation antipsychotics for any atypical examined. These findings are substantially different from other reports.