Numerical studies have been conducted for three arbitrarily chosen engineering problems to compare the interpolating performance of the artificial neural networks with that of the existing regression models. In two example problems modeling nonlinear dynamic processes, the regression models combined with the recursive least-squares estimation showed much better performance in interpolating accuracy as well as convergence rate. In the mapping problem of a PVT-diagram of superheated steam, the neural net-works failed to converge. As far as nonlinear modeling problems are concerned, the conventional regression models still seem to work better, whereas the artificial neural networks have to be improved more before they can play as generic self-learning modelers.