In the paper, the continuous hidden Markov model is applied in static gesture recognition for the first time. According to the characteristics of CHMM, the pixel coordinates sequence of hand contour is chosen as a learning sample. It extracts and tracks the hand gesture by Kinect, and trains CHMM model library which is used for static gesture recognition. At last, the method is compared with the SVM method in an experiment. Experimental result shows the method is efficient, flexible and need minority samples.