We have derived a fall detection algorithm with high sensitivity and specificity from a single accelerometer device worn at the hip. A small clinical trial to obtain accelerometer data corresponding with actual falls experienced by elderly patients failed to provide a statistically significant number of fall events from which to develop an algorithm. Consequently, the detection algorithm was based on analysis of acceleration data containing 201 simulated falls. Although simulated, falls were modelled on video data of actual falls recorded in an elderly population. Nineteen different fall types were represented in the simulated data set which is advancement on previous simulation studies.