Aiming at solving lacking of failure data and inefficiency,high-cost of now available fault diagnosis methods,a experimental platform of fuel transfer system is developed and a fault diagnosis method based on wavelet packet analysis and improved support vector machine（ ISVM） is presented. Based on the vibration signal and the outlet pressure signal obtained from the airborne fuel oil system,the energy of different frequency bands of vibration signal extracted by wavelet packet decomposition can be regarded as characteristic parameters to structure fault feature vector as well as the mean outlet pressure. The genetic algorithm is presented to optimize the parameters of SVM. Meanwhile,the fault feature vectors are used to train and validate this classification model. The experimental results show that this method not only can improve the classification accuracy on fault diagnosis,but also can optimize the use of sensors by using only one vibration sensor and one pressure sensor to recognize the fault signals.