We undertook a randomised controlled trial to assess whether a music therapy (MT) scheme of administration, including three working cycles of one month spaced out by one month of no treatment, is effective to reduce behavioural disturbances in severely demented patients. Sixty persons with severe dementia (30 in the experimental and 30 in the control group) were enrolled. Baseline multidimensional assessment included demographics, Mini Mental State Examination (MMSE), Barthel Index and Neuropsychiatry Inventory (NPI) for all patients. All the patients of the experimental and control groups received standard care (educational and entertainment activities). In addition, the experimental group received three cycles of 12 active MT sessions each, three times a week. Each 30-min session included a group of three patients. Every cycle of treatment was followed by one month of wash-out. At the end of this study, MT treatment resulted to be more effective than standard care to reduce behavioural disorders. We observed a significant reduction over time in the NPI global scores in both groups (F(7,357) = 9.06, p < 0.001) and a significant difference between groups (F(1,51) = 4.84, p < 0.05) due to a higher reduction of behavioural disturbances in the experimental group at the end of the treatment (Cohen's d = 0.63). The analysis of single NPI items shows that delusions, agitation and apathy significantly improved in the experimental, but not in the control group. This study suggests the effectiveness of MT approach with working cycles in reducing behavioural disorders of severely demented patients.