Purpose: The purpose of this paper is to explore how raters combine constituent components of Historical Clinical Risk-20 (HCR-20) risk assessment, and how relevant they rate the tool to different diagnostic and demographic groups. Design/methodology/approach: A cross-sectional survey design of n=45 mental health clinicians (psychiatrists, psychologists, and others) working in a secure hospital responded to an online survey about their risk assessment practice. Findings: HCR-20 Historical and Clinical subscales were rated the most relevant to violence prediction but four of the five items rated most relevant were Historical items. A recent history of violence was rated more important for risk formulation than Historical and Risk management items, but not more important than Clinical items. While almost all respondents believed predictive accuracy would differ by gender, the tool was rated similarly in terms of its relevance for their client group by people working with men and women, respectively. Research limitations/implications: This was an exploratory survey and results should be verified using larger samples. Practical implications: Clinicians judge recent violence and Clinical items most important in inpatient violence risk assessment but may overvalue historical factors. They believe that recent violent behaviour is important in risk formulation; however, while recent violence is an important predictor of future violence, the role it should play in SPJ schemes is poorly codified. Social implications: It is important that risk assessment is accurate in order to both protect the public and to protect patients from overly lengthy and restrictive detention. Originality/value: Despite the vast number of studies examining the predictive validity of tools like HCR-20 very little research has examined the actual processes and decision-making behind formulation in clinical practice.