Background Violent and non‐violent sexual behaviour is a fairly common problem among secure mental health service patients, but specialist sexual violence risk assessment is time‐consuming and so performed infrequently. Aims We aimed to establish whether a commonly used violence risk assessment tool, the Health Clinical Risk management 20(HCR‐20), has predictive validity specifically for inappropriate sexual behaviour. Methods A pseudo‐prospective cohort design was used for a study in the adult wards of a large provider of specialist secure mental health services. Routine clinical team HCR‐20 assessments were extracted from records, and incidents involving inappropriate sexual behaviour were recorded for the 3 months following assessment. Results Of 613 patients, 104 (17%) had engaged in at least one inappropriate sexual behaviour; in 65 (10.6%), the sexual act was violent. HCR‐20 total score, clinical and risk management subscales, predicted violent and non‐violent sexual behaviour. The negative predictive value of the HCR‐20 for inappropriate sexual behaviour was over 90%. Conclusions Prediction of violent sexual behaviour may be regarded as well within the scope of the HCR‐20 as a structured professional judgement tool to aid violence risk prediction, but we found that it also predicts behaviours that may be of concern but fall below the violence threshold. High negative predictive values suggest that HCR‐20 scores may have some utility for screening out patients who do not require more specialist assessment for inappropriate sexual behaviour. Copyright © 2015 John Wiley & Sons, Ltd.