OBJECTIVE: To evaluate the efficacy and safety of low-dose risperidone in treating psychosis of Alzheimer's disease (AD) and mixed dementia (MD) in a subset of nursing-home residents who had dementia and aggression and who were participating in a randomized placebo-controlled trial of risperidone for aggression. METHOD: This post-hoc analysis included only patients diagnosed with AD or MD with psychosis, defined by a score of >or= 2 on any item of the Behavioral Pathology of Alzheimer's Disease (BEHAVE-AD) psychosis subscale at both screening and baseline. Co-primary efficacy endpoints were changes in scores on BEHAVE-AD psychosis subscale and Clinical Global Impression of Change (CGI-C). RESULTS: Overall, 93 patients (46 risperidone and 47 placebo) fulfilled the psychosis of AD criteria. Mean change at endpoint in BEHAVE-AD psychosis subscale with risperidone was superior to placebo (-5.2 vs -3.3; p = 0.039). Distribution of CGI-C at endpoint also favoured risperidone (p < 0.001). The superior improvement with risperidone compared with placebo occurred as early as the first two weeks and persisted to the end of the treatment period. At endpoint, 59% of risperidone-treated patients were responders (i.e. were 'very much' or 'much' improved) compared with 26% of patients receiving placebo. The mean risperidone dose was 1.03 +/- 0.61 mg/day. Twelve weeks of treatment were completed by 37 patients treated with risperidone (80%) and 35 with placebo (74%). A total of 46 (98%) placebo- and 44 (96%) risperidone-treated patients experienced at least one adverse event, with only somnolence occurring more frequently in the risperidone group. CONCLUSION: Risperidone effectively reduces psychosis and improves global functioning in elderly patients with moderate-to-severe psychosis of AD and MD. Copyright (c) 2005 John Wiley & Sons, Ltd.