Aiming at the deception phenomena in the electric power market that disorder the power trading, the early warning problem of power users' credit risk in the power market was studied. To solve the problem, an early warning model based on SVM (Support Vector Machine) was purposed. First the evaluation criteria system and grading standard were discussed in detail. Secondly the early warning model based on SVM was built and the optimization of parameters was analyzed. Finally the data of Liaoning Electric Power Co., Ltd., from 2013 to 2016 is chosen as the sample data for verifying and simulating the early-warming model of power customers' credit risk. The results showed the effectiveness of the proposed method.