BACKGROUND: Benzodiazepines have many benefits for persons with severe mental disorders, but they may also lead to or exacerbate substance abuse. An American Psychiatric Association taskforce established practice guidelines in 1990 to assist physicians in managing these and other potential side effects of benzodiazepine use. The objectives of this study were to determine the prevalence of benzodiazepine use among persons with psychiatric disorders and to evaluate compliance with published prescribing guidelines. METHOD: We studied benzodiazepine use among New Hampshire Medicaid beneficiaries aged 18 to 64 years with ICD-9 diagnoses that were grouped under the headings "schizophrenia," "bipolar disorder," "major depression," and "other psychiatric disorders" from Jan. 1995 through Dec. 1999. Rates and length of use, frequency of high-potency/fast-acting prescriptions, and diazepam-equivalent dosages were compared for those with and without retrospectively determined evidence of substance abuse, substance dependence, or a procedure code indicating treatment for a substance use disorder (SUD). RESULTS: Five-year prevalence of benzodiazepine use for persons with and without SUD was 63% versus 54% for schizophrenia, 75% versus 58% for bipolar disorder, 66% versus 49% for major depression, and 48% versus 40% for other psychiatric disorders. Differences were statistically significant over 5 years and in 1999 (p <.0001). Among persons with major depression or other psychiatric disorders, those with comorbid SUD were more likely to use fast-acting/high-potency benzodiazepines; there were no such differences for those with schizophrenia or bipolar disorder. Persons with bipolar disorder or other psychiatric disorders and SUD received significantly higher diazepam-equivalent dosages than did those without SUD. CONCLUSION: Contrary to published guidelines, rates of benzodiazepine use are higher among Medicaid beneficiaries with severe mental illness and co-occurring SUD than among persons with severe mental illness alone. Additional research and possibly a reassessment of prescribing guidelines are recommended.