This paper presents a human pointing action recogniz-ing system called Finger-Pointer. This system recognizes pointing actions and simple hand forms in real-time by an image sequence processing of stereoscopic TV cameras. The operator does not need to wear any special devices such as Data-Glove. Fast image processing algorithms employed in this system enable real-time processing on a graphic work-station without any special image processing hardware. The system can detect stable and accurate pointing regardless of the operator's pointing style.