Risk factors for human-directed aggression were investigated using retrospective analysis of data from a referral-level clinical behaviour population in the UK. A sample of 200 cases involving human-directed canine aggression and 200 control cases involving no instance of human-directed aggression were selected at random from a population of 746 cases. The final model suggested that clinical cases with human-directed aggression were significantly younger than those presenting with other undesired behaviours (P=0.008) and that male dogs were 1.4 times more likely to be aggressive towards human beings than female dogs (P=0.019). Dogs were 1.7 times more likely to be aggressive towards people if they had attended more than five puppy classes than if they had never attended puppy class (P=0.015) and that dogs were 2.8 times more likely to be aggressive towards human beings if there was another dog between 0 months and 24 months of age in the home (P=0.004). These factors only account for 7 per cent to 10 per cent of the variance between the human-directed aggression population and the control population, but factors such as attendance at puppy classes and numbers of dogs in the household suggest the need for longitudinal studies to investigate temporal relationships.