Abstract: Canonical correlation analysis (CCA) is a classic feature extraction method and is widely applied in pattern recognition. But in face recognition and other small sample size (SSS) problem, its typical disadvantages are: 1) CCA fails, if directly applied, due to the singularity of the covariance matrices of its two groups of features caused by the SSS problem; 2) it can not describe the nonlinear face recognition problem well, for its globally linear property in nature; 3) it is short of the robustness to local variants. Enlightened by our previous sub-pattern PCA (SpPCA) we present sub-pattern canonical correlation analysis (SpCCA) in this paper. By maximizing the correlation between the local and global features of the original samples, this method can not only fuse local and global features well but also eliminate the redundant information among the features. By combining with the sub-pattern method, SpCCA avoids the SSS problem, realizes the formulation for the nonlinear face recognition problem better, and enhances the robustness to the local variants by voting. Experiments on AR and Yale face databases show that the proposed method is stable, robust, and effective.