In this paper, we propose a new system for isolated sign language recognition (SLR) and continuous SLR. In isolated SLR, Histogram of Oriented Displacement is used to describe the trajectories, and multi-SVM is adopted for classification. In continuous SLR, we propose a Dynamic Programming method with warping templates obtained by Dynamic Time Warping (DTW) algorithm. We evaluate our approach with 450 phrases and 180 sentences recorded by Kinect and compare with classical methods, including Hidden Markov Models and state-of-the-art Conditional Random Fields (CRF), Hidden CRF and Latent Dynamic CRF. The experiments demonstrate the effectiveness of our method.