Abstract: A smile expression classification system on data sets of GENKI is built which can represent real-world environments, and the support vector machine and GentleBoost algorithm are used as tools to learn. The followings are introduced: data preprocessing, Gabor features extraction, PHOG features extraction, and local binary pattern features extraction. The experiment results and detailed analysis of these results are given. The empirical study on the GENKI dataset shows the effectiveness of the methods in this article.