Background The quality of care for people with dementia in care homes is of concern. Interventions that can improve care outcomes are required. Objective To investigate the clinical effectiveness and cost-effectiveness of Dementia Care Mapping™ (DCM) for reducing agitation and improving care outcomes for people living with dementia in care homes, versus usual care. Design A pragmatic, cluster randomised controlled trial with an open-cohort design, follow-up at 6 and 16 months, integrated cost-effectiveness analysis and process evaluation. Clusters were not blinded to allocation. The primary end point was completed by staff proxy and independent assessors. Setting Stratified randomisation of 50 care homes to the intervention and control groups on a 3 : 2 ratio by type, size, staff exposure to dementia training and recruiting hub. Participants Fifty care homes were randomised (intervention, n = 31; control, n = 19), with 726 residents recruited at baseline and a further 261 recruited after 16 months. Care homes were eligible if they recruited a minimum of 10 residents, were not subject to improvement notices, had not used DCM in the previous 18 months and were not participating in conflicting research. Residents were eligible if they lived there permanently, had a formal diagnosis of dementia or a score of 4+ on the Functional Assessment Staging Test of Alzheimer’s Disease, were proficient in English and were not terminally ill or permanently cared for in bed. All homes were audited on the delivery of dementia and person-centred care awareness training. Those not reaching a minimum standard were provided training ahead of randomisation. Eighteen homes took part in the process evaluation. Intervention Two staff members from each intervention home were trained to use DCM and were asked to carry out three DCM cycles; the first was supported by an external expert. Main outcome measures The primary outcome was agitation (Cohen-Mansfield Agitation Inventory), measured at 16 months. Secondary outcomes included resident behaviours and quality of life. Results There were 675 residents in the final analysis (intervention, n = 388; control, n = 287). There was no evidence of a difference in agitation levels between the treatment arms. The adjusted mean difference in Cohen-Mansfield Agitation Inventory score was –2.11 points, being lower in the intervention group than in the control (95% confidence interval –4.66 to 0.44; p = 0.104; adjusted intracluster correlation coefficient: control = 0, intervention = 0.001). The sensitivity analyses results supported the primary analysis. No differences were detected in any of the secondary outcomes. The health economic analyses indicated that DCM was not cost-effective. Intervention adherence was problematic; only 26% of homes completed more than their first DCM cycle. Impacts, barriers to and facilitators of DCM implementation were identified. Limitations The primary completion of resident outcomes was by staff proxy, owing to self-report difficulties for residents with advanced dementia. Clusters were not blinded to allocation, although supportive analyses suggested that any reporting bias was not clinically important. Conclusions There was no benefit of DCM over control for any outcomes. The implementation of DCM by care home staff was suboptimal compared with the protocol in the majority of homes. Future work Alternative models of DCM implementation should be considered that do not rely solely on leadership by care home staff. Trial registration Current Controlled Trials ISRCTN82288852. Funding This project was funded by the National Institute for Health Research (NIHR) Health Technology Assessment programme and will be published in full in Health Technology Assessment; Vol. 24, No. 16. See the NIHR Journals Library website for further project information.