The Historical, Clinical and Risk Management Scales (HCR-20) may be a better predictor of inpatient aggression for selected demographic and clinical groups but homogeneity of study samples has prevented definitive conclusions. The aim of this study, therefore, was to test the predictive validity of the HCR-20 as a function of gender, diagnosis, age, and ethnicity while controlling for potential covariates. A pseudo-prospective cohort study (n=505) was conducted in a UK secure/forensic mental health setting using routinely collected data. The HCR-20 predicted aggression better for women than men, and for people with schizophrenia and/or personality disorder than for other diagnostic groups. In women, the presence of the risk management items (R5) was important while men's aggression was best predicted solely by current clinical features from the C5 scale. R5 items were better than C5 items for predicting aggression in people with organic and developmental diagnoses. Our data provide additional information on which HCR-20 raters can formulate overall summary judgements about risk for inpatient aggression based on important demographic and clinical characteristics.