A nondestructive testing method, called spectral analysis of surface waves, was used to examine the behavior of multi-layer cement mortar slab systems under the finite boundary conditions and the different layer properties. General procedure of this test is to generate, measure, and process the dispersive surface waves. After processing the signals, surface wave velocities can be obtained by constructing an experimental compact dispersion curve, which is a plot of surface wave velocity versus wavelength. Surface wave velocities also can be obtained by constructing a theoretical dispersion curve. The values of surface wave velocities obtained from the theoretical dispersion curves were found to have lower values than the results obtained from the experimental compact dispersion curves due to different boundary conditions and reflections from the boundaries. Knowing the difference, the actual surface wave velocity can be obtained considering the results. Using a forward modeling when theoretical dispersion curve is constructed, the shear wave velocities can be found, which can be related to various material properties. Among the signal processing procedure, conventional forward modeling process to obtain shear wave velocity profile has many drawbacks. This study presents the improvement of forward modeling solutions numerically using neural networks. The training sets were prepared using the training pairs obtained from the actual experimental results and computed velocities. The training sets were then trained using the back-propagation training algorithms. Test runs were conducted, which showed a close agreement with the experimental results.