A hydraulic fault diagnosis method that combined sensitivity analysis and probability neural network (PNN) is proposed to improve the speed and accuracy of the hydraulic pump fault diagnosis. First, the timedomain figure and the spectrum figure that under various states were analyzed to find using the traditional method to diagnose the fault of the hydraulic pump is difficult. Then, the feature parameters that under various states were extracted, and the sensitivity of the feature parameters were analyzed. Finally，constituted vectors with feature parameters that have the higher sensitivity, used the vectors to train PNN, and used the trained PNN to diagnose fault of the hydraulic pump. Experiment show that PNN can quickly and accurately diagnose the fault, save the diagnostic time. And combined sensitivity analysis and PNN can improve the correctness of PNN.