Benzodiazepines are frequently long-term prescribed for the treatment of patients with severe mental illness. This prescribing practice is problematic because of well-described side effects including risk of dependence. We examined the efficacy of prolonged-release melatonin on objective and subjective sleep quality during benzodiazepine discontinuation and whether sleep variables were associated with benzodiazepine withdrawal. Eligible patients included adults with a diagnosis of schizophrenia, schizoaffective disorder, or bipolar disorder and long-term use of benzodiazepines in combination with antipsychotics. All participants gradually tapered the use of benzodiazepines after randomization to add-on treatment with melatonin versus placebo. Here we report a subsample of 23 patients undergoing sleep recordings (one-night polysomnography) and 55 patients participating in subjective sleep quality ratings. Melatonin had no effect on objective sleep efficiency, but significantly improved self-reported sleep quality. Reduced benzodiazepine dosage at the 24-week follow-up was associated with a significantly decreased proportion of stage 2 sleep. These results indicate that prolonged-release melatonin has some efficacy for self-reported sleep quality after gradual benzodiazepine dose reduction, and that benzodiazepine discontinuation is not associated with rebound insomnia in medicated patients with severe mental illness. However, these findings were limited by a small sample size and a low retention rate.