Objective: To evaluate the effect of a musical exercise programme on mood state and cognitive function in women with dementia. Design: Randomized controlled trial. Setting: Public Psychiatric Hospital Rekem, Belgium. Patients: Twenty-five patients with dementia. Interventions: Fifteen patients attended exercise training for three months, which consisted of daily physical exercises supported by music for 30 min/session. They were compared with a group of 10 control patients, who received an equal amount of attention through daily conversation. Main measures: The effect on cognition was measured by the Mini-Mental State Examination (MMSE) and the Amsterdam Dementia Screening Test 6 (ADS 6). Behaviour was evaluated with the abbreviated Stockton Geriatric Rating Scale (BOP scale). The assessments were made before, after six weeks of intervention and immediately after the three-month experimental period. Results: The exercise group showed a significant improvement in cognition. This was documented by an increased MMSE mean score of 12.87-15.53, and by a higher median score, rising from 10 to 14 points, on the subset 'fluency' (ADS 6 test). The control group showed no significant improvement, either on the MMSE (mean score of 10.80-11.00) or on the fluency subtest of the ADS 6 (median scores were 6.5-7 points). The effects on behavioural changes were not significant. Conclusion: The present study suggests a beneficial effect of cognition using a music-based exercise programme in a group of patients with moderate to severe dementia. Further studies are needed to confirm these findings.