Abstract: Both of the experiments and numerical simulations of damage detection for timber beams with cracks based on electromechanical impedance method were conducted. The BP neural network was used to classify and identify the location of timber beam damage. Firstly, the influence of the cracks at different depths and different positions on impedance was experimentally tested. Then the damage index, root mean square deviation (RMSD), was employed quantitatively to evaluate the relationship between the impedance values and the damage severity and location of the timber beams. Both of the experimental and numerical results indicate that the corresponding frequency at the peak impedance decreased slightly when the crack appeared in timber beam. In addition, the RMSD increased gradually with the increase of the damage severity, however, it decreased gradually with the increase of the distance between the piezoelectric sensors and crack location.