Abstract: Swelling potential of expansive soil is controlled by many factors. So quantitative classification of expansive soil is always a problem and is not solved well. This paper takes the expansive soil in the Water Transfer Project from the Yangtze River to Han River in middle scheme of the South-to-North Water Diversion Project as the research object. Through analysis of experimental datas, this paper gives the suggested value of classification standard of the expansive soil. Based on three mathematical methods of artificial neural network model, the Fisher's discriminant analysis and rough sets theory, comprehensive indexes are used to analyze the swelling potential of 40 expansive soil samples. This paper determines the final classification results through comparative study of the three mathematical methods. According to the classification results and operation method of each mathematical method, based on the theory of relatively simple, easy to operate and easy to accept, this paper suggests that the Fisher's discriminant analysis can be a priority to analyze the swelling potential of expansive soil. In addition, this paper also gives the suggested value of classification standard of montmorillonite content, which verifies the correctness of classification results. The results of this paper provide a theoretical basis for the identification and classification of expansive soil in the process of the rest of excavation.