An improved face recognition method based on Fisher linear discriminant analysis is proposed,which can deal with the problem that the class mean of training samples deviates from the center of this class in small sample size case.The feature vectors are resulted from Karhunen-Loeve(K-L) transformation,and the weight of each vector is obtained according to the distance between the feature vector and the others in the same class.After that,the new class means are calculated using the weighted feature vectors.And then the within-class scatter matrix and between-class scatter matrix are rebuilt,which improve the Fisher linear discriminant function.The experiments on ORL and Yale face databases show that the proposed method is superior to principe component analysis(PCA) and fisherface methods.