OBJECTIVE: To examine the predictive validity of unstructured clinical risk assessment and associated risk factors for aggression in predicting self- and other-directed aggression in the first 4 weeks of admission for patients admitted to an Australian adolescent psychiatric inpatient facility. METHOD: A retrospective review of patient records was conducted at the Marian Drummond Adolescent Unit during late 2009 for the period of September 2006 to July 2009. Information collected included admission risk assessment ratings, aggressive incident reports, patient diagnoses, sex and history of aggression and self-harming behaviour. RESULTS: A total of 193 adolescents (aged 13-18 years old) were included in retrospective analyses. The hypothesis that unstructured clinical risk assessment would be predictive of self- and other-directed aggression was partially supported. High risk assessment scores were predictive of engagement in other-directed aggression. A history of physical aggression was also found to be predictive of engagement in other-directed aggression; however, it was not as predictive as the risk assessment rating. High risk assessment scores were not predictive of self-directed aggression. A history of engaging in one or more acts of self-harm or suicide was the most predictive of engagement in self-directed aggression during inpatient stay. Female sex also predicted engagement in self-directed aggression. CONCLUSIONS: Based on professional expertise, prior experience and intuition, clinicians are relatively good predictors of other-directed aggression in adolescent inpatient units; however, they are less successful at predicting self-directed aggression in this population. It is possible that, unlike other-directed aggression, self-harming behaviour is heavily dependent on environmental factors and that admission to the inpatient unit removes these triggers from the individual's environment.