Abstract: Under the normal circumstances in the laboratory,a series of testing for detecting and diagnosing,both qualitatively and quantitatively,and the defects of holes with different diameters in 50 maple wood samples using RSM-SYS ultrasonic instrument were carried out.The 32 coefficients of wavelet packet from low frequency to high frequency were obtained by wavelet packet analysis of eigenvalues derived from ultrasonic signals by using Matlab.The most important eigenvalue,energy variation sequences of a signal at different frequency bands,was extracted from the coefficients and a new 32 dimension eigenvector was composed of those eigenvalues,which can be used as the input matrix for the Back Propagation Artificial Neural Network(BP ANN) for learning,training and diagnosing.The results showed that the accuracy rate of detecting and diagnosing hole defects in A.mono wood was up to 88% and the index of correlation between simulation outputs and test objective outputs of BP ANN diagnosing system was ranging from 0.8 to 0.9.