Human gesture is commonly used in daily communication between humans, and there are more and more studies trying to utilize gestures in human computer interaction. Because the advance of wireless and accelerometer technologies, accelerometer which is constrained by the environment compared with other sensing devices is getting more and more interested in these studies. In this thesis, we introduce an accelerometer-based gesture recognition system which is composed of gesture characteristic database and continuous gesture recognition. In the gesture characteristic database, we propose a method to construct a gesture characteristic database which can handle intra-class variations of human hand gestures. In the continuous gesture recognition part, a kernel based matching method and a dynamic time warping based method are proposed to recognize gestures in human hand motions. We here design two different gesture sets in the experiments, and the experiment results show that our system can recognize continuous hand gestures quickly and accurately.