Background Preliminary studies have indicated that training staff in Positive Behaviour Support (PBS) may help to reduce challenging behaviour among people with intellectual disability (ID). Objective To evaluate whether or not such training is clinically effective in reducing challenging behaviour in routine care. The study also included longer-term follow-up (approximately 36 months). Design A multicentre, single-blind, two-arm, parallel-cluster randomised controlled trial. The unit of randomisation was the community ID service using an independent web-based randomisation system and random permuted blocks on a 1 : 1 allocation stratified by a staff-to-patient ratio for each cluster. Setting Community ID services in England. Participants Adults (aged > 18 years) across the range of ID with challenging behaviour [≥ 15 Aberrant Behaviour Checklist – Community total score (ABC-C_T)]. Interventions Manual-assisted face-to-face PBS training to therapists and treatment as usual (TAU) compared with TAU only in the control arm. Main outcome measures Carer-reported changes in challenging behaviour as measured by the ABC-C_T over 12 months. Secondary outcomes included psychopathology, community participation, family and paid carer burden, family carer psychopathology, costs of care and quality-adjusted life-years (QALYs). Data on main outcome, service use and health-related quality of life were collected for the 36-month follow-up. Results A total of 246 participants were recruited from 23 teams, of whom 109 were in the intervention arm (11 teams) and 137 were in the control arm (12 teams). The difference in ABC-C_T between the intervention and control arms [mean difference –2.14, 95% confidence interval (CI) –8.79 to 4.51; p = 0.528] was not statistically significant. No treatment effects were found for any of the secondary outcomes. The mean cost per participant in the intervention arm was £1201. Over 12 months, there was a difference in QALYs of 0.076 in favour of the intervention (95% CI 0.011 to 0.140 QALYs) and a 60% chance that the intervention is cost-effective compared with TAU from a health and social care cost perspective at the threshold of £20,000 per QALY gained. Twenty-nine participants experienced 45 serious adverse events (intervention arm, n = 19; control arm, n = 26). PBS plans were available for 33 participants. An independent assessment of the quality of these plans found that all were less than optimal. Forty-six qualitative interviews were conducted with service users, family carers, paid carers and service managers as part of the process evaluation. Service users reported that they had learned to manage difficult situations and had gained new skills, and carers reported a positive relationship with therapists. At 36 months’ follow-up (n = 184), the mean ABC-C_T difference between arms was not significant (–3.70, 95% CI –9.25 to 1.85; p = 0.191). The initial cost-effectiveness of the intervention dissipated over time. Limitations The main limitations were low treatment fidelity and reach of the intervention. Conclusions Findings from the main study and the naturalistic follow-up suggest that staff training in PBS as delivered in this study is insufficient to achieve significant clinical gains beyond TAU in community ID services. Although there is an indication that training in PBS is potentially cost-effective, this is not maintained in the longer term. There is increased scope to develop new approaches to challenging behaviour as well as optimising the delivery of PBS in routine clinical practice. Trial registration This study is registered as NCT01680276. 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. 22, No. 15. See the NIHR Journals Library website for further project information.