Objective: The present study evaluates the accuracy of clinical and archival predictors of patients' aggressive behaviour on a locked admissions ward. Method: Over a 9-month period, staff members estimated the likelihood that patients would become aggressive during their stay in the ward. These unaided clinical assessments were obtained with Visual Analogue Scales (VASs) administered before the end of the first full day of admission. Archival predictions were based on demographic variables (e.g. gender, number of previous admissions, diagnosis) derived from patients' admission forms. Aggressive behaviour was recorded with the Staff Observation Aggression Scale-Revised (SOAS-R). Results: Clinical predictions of aggression were found to be moderately accurate. On the basis of clinical estimates, 75% of the patients were correctly classified as becoming aggressive or not. Conclusion: Although a body of evidence indicates that unaided clinical prediction of violent recidivism after hospital discharge does not perform well, it may be quite accurate in estimating short-term aggression risks during acute psychiatric admission.