Abstract: An inverse method is presented to determine the elastic constants of an experimental sample, a titanium graphite unidirectional fiber-reinforced composite plate, using wavelet transform and neural networks. Optimal algorithms of wavelet transform and neural networks are given here in order to improve the accuracy of inversion results. Coherent results were shown in both fiber direction and cross fiber direction, proving the feasibility of this method. Neither the group velocity of the Lamb wave modes are needed, as in the conventional method, and no direct least-square fitting of the experimental waveforms is necessary.