Although many different behavior recognition algorithms have been prop osed,most of them are applied into RGB video sequences.What is worse,their performance is not satisfactory .Thus,in this paper a behavior recognition algorithm based on Kinect and pyramid features is proposed,in which Kinect is able to obtain RGB information,and the corresponding depth information is very help ful for our task.What is more, pyramid features not only describe the global shape and the local details of hu man behavior,but also depict the spatial information of human behavior.The combination of multi-kernel support v ector machine (SVM) classifier in challenging DHA dataset experimental results show that the performance of pyramid features on bo th RGB and the depth channels is satisfactory,and when these features are fu sed,its recognition rate is further improved to 96.2%,whose performance is much better than that of the state of the art behavior descriptor s.