In view of the aero engine sensor fault phenomena, combined with sparse support vector machines and robustness, aero engine sensor fault diagnosis is designed by using SVM. SVM is trained out of line, and used on line. Compared the output results with the actual system output, it can produce high precision fault residuals by the simulation system's dynamic characteristics which was having been trained in accordance with SVM as the core, through the residuals to determine sensor failure. The simulation results show that the method in the aviation engine sensor fault diagnosis can be better to simulate the dynamic characteristics of the tested system, and can timely and accurately locate faults.