Based on the principal component analysis, 1ocality preserving projection and kernel mapping theory, kernel within-class neighborhood preserving orthogonal algorithm with global structure is proposed. The object function ofproposed method takes the global and local information of the samples into account. At the same time, by using nonlinear mapping and the orthogonal restriction of base vectors, this algorithm can extract more effective classification features. Being unknown nonlinear function, the object function can not be directly calculated. But according to the kernel mapping theory, the equivalent object function of the proposed algorithm is established in the kernel feature space. The deducing process and solving steps are given. Experimental results on face database demonstrate the effectiveness of the proposed method.