A method used to recognize the inner defects of 3-D braided composite materials is discussed. Firstly, the link between UT signals and the defects of 3-D braided composite material is analyzed. Then, the wavelet packet transform is used to process the ultrasonic scanning pulse signals of the defects. The characteristic quantities of signal are extracted into the BP neural network as samples. Through training the BP neural network, the recognize system of micro-cracks and pores is achieved. Finally, according to the results of experiment this classification system based on wavelet packet transform is proved to be feasible.