Abstract: Sousachinensis is the country grade one protected animal in China, previous researches have proved that there are differences in the echolocation (click) signals of sousa chinensis in different sea areas. In this paper, three machine learning methods (K-Nearest Neighbor, Decision Trees-Classification and Regression Tree, Navie Bays) are used to identify Sousachinensis on the basis of echolocation signals. Firstly, the Teager-Kaiser Energy Operator (TKEO) method and Gabor filter are used to automatically detect the click signal of Sousachinensis, then the cepstrum operation is applied to the echolocation signal (click) for a better feature representation before using the three machine learning methods. Experimental researches are conducted on the signals colleted from the Sousachinensis in Leizhou Bay and the Sousachinensis and bottlenose dolphins in Xiamen sea area. In the experiment, the Sousachinensis in Xiamen sea area and the the Sousachinensis in Leizhou bay are set as the first group to indentify the same species of dolphin; and the the Sousachinensis and the bottlenose dolphin in Xiamen sea area are set as the second group to identify different species of dolphins. The results show that the average recognition accuracy of these two groups of experiments can reach 98% and 96%, respectively.