To help solve the dimension reduction of the ault features in analog circuits, a method called local linear discrimination analysis (LLDA) is proposed by introducing the idea of localization in linear discrimination analysis (LDA). A local data patch of a single group of data is constructed, and the scatter matrices of the intra-class and inter-class are computed. Next, the local optimization criterion is established by balancing the difference between the intra-class compactness and inter-class separation by the scaling factor. Then, the objective function is reconstructed on the alignment algorithm in the whole data space. Finally, the projection matrix is obtained by conducting the standard eigenvalue decomposition for dimension reduction. The algorithm makes full use of the local discrimination information to solve the nonlinearity of the data while maintaining the class differentiation, and avoids the problem of small samples in LDA. Studies on the standard dataset and extraction of fault features in the analog circuits demonstrate the effectiveness of LLDA.