In this paper, artificial neural network was used for estimating crack depth in aluminum plates. Finite element simulations were performed and the results showed that the transmitted S0 mode wave decreases with the in-creasing of crack depth. The amplitude and the energy of the S0 wave showed good correlation with crack depth, so they were used as features for neural network training. Finally, experiments were performed with an aluminum plate with three cracks. The features of the experimentally obtained signals were used for testing the performance of the network. The crack depth estimation results showed a maximum relative error of 8%.