The aim of this study was to examine the risk of injury among patients and staff following involvement in a restraint episode in relation to restraint position (standing, supine or prone) and other aspects of the pre-incident behaviour including perceived causation. Mixed effects logistic regression was used to estimate the relative odds of injury to staff or patient in a series of 680 restraint episodes involving 260 patients in an adult mental health service in England between 1999 and 2001. There was no statistically significant association between patient injury and restraint position in this sample, but a prone restraint position was weakly associated with staff injury. Staff injury was most likely when an actual assault had occurred prior to the incident. Patient injury was more likely when the patient had self-harmed, had been abusing substances and had used a weapon prior to the incident, and less likely when the patient was showing signs of frustration with their environment. The use of prone restraint may be weakly associated with an increased risk of injury to staff. However, other aspects of the incident are stronger predictors and should be considered when planning training for front line staff.