Background The HCR‐20 is a widely used 20 item structured professional judgement aid to risk assessment and management, but longitudinal studies of its value are rare, particularly with people at high risk of reoffending. Aims To investigate whether the HCR‐20 discriminates between patient subgroups in one high‐security hospital in England, whether scores reduce with hospital treatment and whether lower scores predict discharge. Methods Repeated HCR‐20 ratings were made by clinical teams across five services within the hospital, two of them (women and men with intellectual disability) national services. A database of 3337 HCR‐20 ratings, from 532 patients over a period of 5 years, was examined using mixed effects models. Results As expected, HCR‐20 scores were high overall, but there were differences between services in the ratings obtained. Female patients and men with intellectual disability had the highest total score. There was a significant relationship between discharge and lower clinical risk score, but not between total and risk scale scores and discharge. There were significant changes in scores over time, although these were small and may not be clinically meaningful. Differences between services were observed, with women evidencing greater change. Conclusions It is unsurprising that patients in two national services (for men with intellectual disability and women) have the highest HCR‐20 scores; however, the finding of relatively greater risk reduction in women needs further investigation. Although we did not find ceiling effects in this sample, the clinical value of frequently repeated HCR‐20 ratings may be limited for high‐risk populations where any change is likely to be slow. Copyright © 2013 John Wiley & Sons, Ltd.