The accurate prediction of the cylinder fatigue failure is a key part of Condition Based Maintenance (CBM) to keep the system operating reliably. In additional, it can provide pertinent service value required in the design process for preventing the redundancy design of cylinders. The common used failure prediction method based on failure samples is quite limited as the samples data are always too rare to build a rational prediction model. A new fatigue failure prediction method based on Principal Component Analysis (PCA) and Wavelet Neural Network (WNN) is established, which is used in prediction of cylinder performance failure. By taking full advantages of the realtime performance parameters of the cylinder in operation obtained through processing of PCA method, the parameters were reduced without losing much key information at the best, and then the sample data grouped rationally were input to the WNN, whose initial weights and thresholds were obtained by using Genetic Algorithm (GA). The training and testing procedures of the WNN were realized to complete the predition of cylinder performance failure. The experimental results show that the proposed method has a satisfactory effects on the fatigue failure prediction of cylinders.