BackgroundChild maltreatment has a pervasive, detrimental impact on children's wellbeing. There exist gaps in evidence on school-based prevention programmes—namely, failure to use experimental designs and standardised validated measures in evaluation, and design limitations in terms of their duration, intensity, and breadth. We aimed to address these gaps in evaluating the effectiveness of Keeping Safe, a whole school child maltreatment prevention programme. MethodsIn a two-arm cluster-randomised controlled trial, participating schools were randomised (1:1, wait-list control group) using a computer-generated list. Intervention schools implemented Keeping Safe between Sept 16, 2016, and June 30, 2018, while wait-list control schools continued to teach the statutory curriculum as usual. Keeping Safe, which is premised on a whole-school model with components covering all forms of maltreatment, targets children, parents, and school staff. Power calculations indicated that approximately 40 schools are required in each arm of the study to detect a minimum detectable effect size of 0·19 SD with 80% power and 95% significance. This was to allow for a minimum detectable percentage point difference of 2·7 for an outcome that is rare in the control group, at around 5% (ie, disclosure of abuse). Data were collected by teachers and school staff under classroom test conditions following a standard data collection procedure devised by the research team. It was not possible to blind school staff. However, the outcome data were obtained through a self-report measure reducing potential outcome assessor bias. Data processors and the trial statistician were blinded to allocation. Data were collected at baseline (June, 2016), and 1 and 2 years after programme implementation. Primary outcome measures included Children's Knowledge of Abuse Questionnaire (CKAQ), a 33-item questionnaire designed to assess children's knowledge of sexual abuse, and the “What If” Situations Test (WIST), which assesses children's ability to “Identify”, “Refuse”, “Escape” “Tell”, and “Report” inappropriate situations. Analysis was by intention-to-treat and conducted at the level of the individual, taking into account the clustering effects of schools. Here we report analysis of data collected after the first year of implementation (June, 2016, to June, 2017). Ethics approval was given by the NSPCC Research Ethics Committee R/15/67 Findings33 schools were allocated to the intervention (1649 children) and 31 to control (1550 children) groups. The ITT analysis ensures that children included at baseline were included at Time 1. After year 1 the intervention group had 31 schools (two schools withdrew in July, 2016, after randomisation, citing changing school leadership as the reason for withdrawal) and the control group had 30 schools (one withdrew in June, 2016, citing sensitive content of questions in outcome measures as reason for withdrawal). There were no statistically significant differences between intervention and control at baseline across all outcomes. After 1 year, the intervention group scored significantly higher than the control group on three of the four knowledge dimensions of the CKAQ (appropriate refusal β=0·241, p=0·001; inappropriate touch β=0·850, p<0·0001; and inappropriate touch in familiar relationships β=0·455, p<0·0001); assessment of appropriate touch was not significantly different. The intervention group also scored higher on four of the five skill dimensions of the WIST (“Identify” β=0·675, p=0·007; “Escape” β=0·741, p=0·049; “Tell” β=0·432, p=0·035; and “Report” β=1·006, p=0·03). InterpretationThese results are encouraging and useful for initial planning around scale-up of the Keeping Safe programme. FundingThe Department of Education Northern Ireland (5 year grant, 2013–18). This funder has no role in study design, collection, management, analysis, and interpretation of data, writing of the report, or the decision to submit the report for publication.