OBJECTIVE The use of physical intervention on psychiatric inpatient units continues to be a source of debate and controversy. Some studies and national clinical guidelines have identified particular restraint positions as both dangerous and undesirable. The following study attempts to identify clinical variables that may make physical restraint in a particular position more likely. METHODS A cross-sectional survey design was adopted and data was obtained from a violence and aggression audit form used by the trust. This form has 122 items to be completed by staff within 72 hours of an episode of patient aggression or self-harm. Ten variables were selected for scrutiny on the basis of their potential clinical importance. RESULTS The survey found that prone restraint was significantly associated with others reporting the patient's imminent violence and high-intensity observation after the incident. Supine restraint was significantly associated with the patient being withdrawn and/or refusing to communicate prior to the episode and with a high severity incident rating after the incident. CONCLUSIONS If we work on the premise that restraint in the prone position is less desirable than interventions undertaken with the patient in the supine position, this study clearly suggests that we have an opportunity to influence the nature of intervention through quite minimal changes to training programmes. It is important that any change in emphasis around intervention does not create a sense that controlled descent to the floor is inevitable. The principle of its use as a 'last resort in the event of loss of control on the feet' has to be maintained.