Accidental falls are common causes of serious injury and health threats in the elder population. To deliver adequate medical support, the robust and immediate falls detection is important. Since the fall detection in the elderly remains a major challenge in the public health domain, effective fall-detection will provide urgent support and dramatically reduce the cost of medical care. In this work, we propose a fall-detecting system placing an accelerometer on the head level and using an algorithm to distinguish between falls and daily activities. The experimental results have demonstrated the proposed scheme with high reliability and sensitivity on fall detection. The system is not only cost effectively but also potable. It fulfills the requirements of fall detection.