In this paper, a complete human face automatic target recognition system is presented. Input images of human faces are grabbed at standard video rates and recognized in real- time. The database contains 200 face images stored in computer memory in advance. The system includes a feature point extractor, a normalization algorithm, an encoding module, an optical recognizer which is based on hit/miss transform and an incoherent correlator, and a postprocessing module. When a real-time face image is picked up by CCD camera, three reference points including two internal corners of both eyes and the mouth center, are extracted automatically for normalizing use. Then the face image is normalized to a fixed scale and posture by using the affine transform. The normalized image will then be encoded to an original-complementary composite encoding image and be sent to a liquid crystal display panel, which is used here as a real-time spatial light modulator, to be recognized by the optical recognizer. Experimental results have shown that the system has an accuracy over 94%. Its tolerance abilities to head turning, nodding, and tilting are 20 degree(s), 5 degree(s), and 15 degree(s) respectively. It is also tolerable to noise disturb up to 30%, to image lost up to 40%, and even to different expressions such as seriousness, smile, and astonishment.