A real-time probablistic face tracking system using monocular vision is presented based on face target acquisition and subsequent particle filtering techniques. First, the face target acquisition and initialization stage used a skin color classification and statistical face model matching approach to find the face target. Subsequently, the particle filtering technique is used to track the state space of face movements. And finally, the optical flow information was used to find motion information for sample redistibution. The system places emphasis on the automatic face target initialization stage, which has been assumed to be solved or labled manually in most other face tracking systems. Using a monocular USB camera on an Intel Pentium III 700 MHz laptop, the face detection and initialization stage is executed in less than 250 msec and the subsequent face tracking stage functions at 30 fps comfortably with 160×120-pixel resolution live videos.