Purpose This study compared the ability of four risk assessment scales to predict non‐sexual violent reoffending, and differences in non‐sexual violent reoffending rates by sexual offending history. Method Risk assessment instruments were scored, and criminal histories and three non‐sexual violent reoffending outcomes were coded, for a large sample of sexual offenders supervised by probation services in England and Wales. Predictive validities for the three outcomes were compared, varying the banding of risk scores to reflect practical constraints on offender management resources. Reoffending rates were compared by sexual offending history. Results After adjusting for risk assessment tool banding, the Offender Group Reconviction Scale version 3 and OASys Violence Predictor (OVP) had generally superior predictive validity to Risk Matrix 2000's v and c scales. However, several of OVP's dynamic risk factors were unrelated to non‐sexual violent recidivism. Non‐sexual violent reoffending rates were greater among those with prior but not current sexual offences and lower among those with indecent images offences, and sexual reoffending rates were lower but not negligible among those who had only sexually offended before the age of 16. Conclusions The use of OVP was recommended to the English and Welsh correctional services. The dynamic risk factor and sexual offence history results suggest that further work is required to optimize prediction of non‐sexual violence among sexual offenders.