A comparison of three different models using back-propagation neural network for estimation of pile bearing capacity from dynamic stress wave data was made. The bearing capacity predicted by TNOWAVE was employed as the desired output in training. The study shows that the neural network models generally predict total bearing capacity more favorably if both the stress wave data and the properties of the driven pile are considered as the input parameters. In addition, better selection of input parameters rather than the increase number of input parameters will improve the accuracy of the prediction.