The electric locomotive obtains the electric energy through the electrical contact between pantograph and catenary. And the arc fault between the pantograph and catenary is a common phenomenon when the electric locomotive is running. The arc fault in pantograph-catenary system will not only damage the catenary and the pantograph, but also disturb the operation of the electric equipment in electric locomotive. The arc fault has the characteristic of randomness and irregularity, so it is difficult to diagnose the fault timely and accurately. Ac- cording to the existing problems in arc fault diagnosis, this paper introduced a new diagnosis method based on the support vector machine(SVM). After obtaining the raw data of the current in pantograph-catenary system, the power spectrum entropy was adopted to extract the feature vectors needed by arc fault diagnosis and the SVM was used to classify these feature vectors. Then, the normal current from arc fault can be recognized. The re- search results show that the diagnosis model established in this paper has a high accuracy in classifying the fault state and normal state, which provides an useful method and research approach for arc fault diagnosis.