Piston head looseness is a typical progressive failure of aero hydraulic pump. It is difficult to make precise fault diagnose because the fault feature is misty, the fault samples are insufficient and the measurable signals are full of structure coupling and noise besides failure feature. In order to solve above problems, a fault diagnosis method based on contracted support vector machine(SVM) was proposed. In the new method, rough set was utilized to reduce the fault characteristic value and eliminate redundancy in order to find the minimal attribute describing system fault characters on the premise of unchanged classification quality. The sample data disposed by rough set were used to train SVM to realize fault diagnosis. The experiment result of the trained contracted SVM shows that this diagnosis method is suitable for the high-precision fault diagnosis of the aero hydraulic pump.