In order to solve low recognition rate and poor robustness in pipe defect recognition based on ultrasonic guided wave, principal component analysis (PCA) is used to optimize the feature of pipe defect echo signals. First, a few characteristic parameters in domain of time and timefrequency were extracted by means of dealing with the ultrasonic guided echo signals of the pipe defect to construct a multifeature vector. The multifeature vector dimension is then reduced using principal component analysis. The fusion feature is generated by extracting the principal component whose cumulative contribution rate is about 89%. Finally, BP neural network is used to train and recognize fusion feature. This method can effectively recognize the pipe defect, and has higher recognition rates than that of the multifeature vector.