Physical restraint of potentially violent patients can take place on the floor in prone or supine positions (horizontal restraint) or, less intrusively, in standing or sitting positions (vertical restraint). The decision to use horizontal restraint may be made unnecessarily if the level of risk in an interaction is overestimated and/or the capacity to otherwise control the situation is underestimated by staff. A model of the decision-making process in this situation based on the concept of cognitive appraisal of imminent threat is proposed here. The aim of the study reported here was to test this model by examining the patient, staff, and environmental factors associated with the use of horizontal (as opposed to vertical) restraint. Logistic regression analysis of secondary data on the first-recorded episode of restraint of 261 patients occurring over a three-year period (1999 - 2001) in an acute care psychiatric trust in the UK was conducted, followed by a ROC analysis of the discriminative ability of the derived model to predict horizontal restraint. It was found that patients admitted under Section 2 of the Mental Health Act, those committing self-harm, and those with unclear thoughts and increased vocal volume prior to the incident were more likely to be restrained horizontally. Older people, those with perceived rational causation such as bad news and personal gain, and those with no obvious warning signs were less likely to be restrained horizontally. The AUC of the model was .76 (95% CI .70 - .82), indicating acceptable discriminative ability. It is concluded that certain patient characteristics and certain staff appraisals of pre-incident behaviour are associated with an increased risk of horizontal restraint.