Fall detection is one of the major issues in health care filed. Falls can cause serious injury both in physiology and psychology, especially to the old people. A reliable fall detector can provide rapid emergency medical care for the fallen down people. Thus, a reliable and effectively fall detection system is necessary. In this thesis, we propose a system which utilizing mobile phones as a detector to detect the falling. When fall accident occurs, the system has three response procedures for help. The first procedure is transmitting the emergency message to the related people for help. The second procedure shows the user’s status and location on the map of webpage, according to user’s GPS location and status. The third procedure makes the alarm sound; its purpose is to let the person who nearby the user can be noticed that the user needs help. First, using a waist-mounted mobile phone to capture accelerometer of the human body and adopt the DCT (Discrete Cosine Transform) to analyze the value of accelerometer to distinguish the activities of daily living (ADL) and falls. ADL consist of walking, standing and sitting. We utilized a tri-axial accelerometer in mobile phone to capture the signal and transmit it to the server by way of Internet. We adapt two judgments achieved in Server, first judgment is based on an adaptive threshold for detecting the energy by DCT; the setting of adaptive threshold include height, weight and gender. The second judgment is according to the tilt of smart phone. Experimental results show that this method can detect the falls effectively; in addition, it is more portable than other devices as well.