Falls often cause serious injury and health threats for elderly people. It is also the major obstacle to independent living for frail and elderly people. Many researchers try to establish an efficient fall prevention strategies for elderly people by collecting a lot of fall characteristics. However, it is difficult to obtain these characteristics simply from the questionnaires of elderly people. Since they may forget or misremember their falling scenario. In this work, we propose a fall characteristics collection system for designing fall prevention strategies. A waist-mounted tri-axial accelerometer is used to capture the movement data of the human body when elderly people fall. Then, the proposed algorithm uses the variations of angle between acceleration vector and three axes to determine the fall characteristics which include falling directions and impact parts. Experimental results demonstrate effectiveness of the proposed scheme. The system is not only cost effective but also portable that fulfills the requirements of fall characteristics data collection.