The human face is a complex pattern. It is very important for recongition to find human faces automatically in a scene. In this paper a new method to locate and segment human faces in a complex background is proposed.This system utilizes a hierarchical knowledge-based method and consists of three leVels. The higher two levels are had on detecting maximum（corropnding ic faces） and minimum （corresponding to eyes） in mutiresolution image using B-pline dyadic wavelet transform

and then we make initial selection using the relation of scale

involvement and symmetry. The last level will determine the position according to the feature of face. This method can locate and segment unknown human faces spanning a wide range of sizes in an intensity image. Some experimental results are also given.