We have been developing a high performance real-time face detector. We built a single strong classifier composed of a long sequence of weak classifiers. A large number of positive training samples and negative training samples are used to decrease generalization error. To classify a large numbers of training samples, replacement of positive and negative training samples for bootstrapping is executed at training each weak classifier. Compared with other state-of-art face detectors, our detector is the most accurate for frontal face detection. The running time of detecting frontal face images for 320 by 240 pixel image is about 20ms. Our face detector can be extended to detect rotated and profile faces.