A friendly and intelligent human-robot interface (HRI) system based on machine vision is established in this paper. A user uses his/her face and hand gestures to communicate with the robot. Three kinds of face detection, including shape, similarity and lip detections, are adopted to improve the recognition accuracy. The recognition of hand gestures is implemented through PCA, BPANN and their combined classifier. Successive dynamic commands are processed by dynamic programming. Experimental results show that the devised natural HRI works well and the system can process 8 image frames per second for practical real time applications.