Abstract: Two methods of extraction of defect features of metal matrix composite by combining multi-resolution signal decomposition of the wavelet packet with energy spectrum were suggested. In the first method, the energy characteristic vector, which is the best parameter that reflects the defect features, is chosen as characteristic parameter for defect recognition. In the other method, it is need to find the most sensitive node for crack defects and analyze its energy spectrum so that one can distinct apparently the signals of crack and non-crack and which one in crack's signals. As an example, the defect signals of steel connected components are processed with two methods It is shown that two methods are effective for extraction and recognition. They develop a new direction for extraction and recognition of the defect of composite material.