In order to detect and recognize the extent of crack damage for piezoelectric flexible cantilever beam, experiments are conducted using the method of combining wavelet packet analysis with wavelet neural network (WNN). The damage index of H2 norm of energy ratio relative variation is given based on wavelet packet energy spectrum. An experimental setup for cracked piezoelectric flexible beam is built up. After excitation, the vibration signals are measured by using two piezoelectric patches and decomposed by using wavelet packet to calculate damage indexes. The damage indexes are combined as input characteristic parameters of WNN to identify the depth of crack in the beam. The experimental results demonstrate that the H2 norm of energy ratio relative variation is sensitive to crack damage of beam, and it is not sensitive to noise. The crack depth of the flexible beam can be recognized by using WNN accurately.