This paper shows the efficiency of neural networks (NN), coupled with the finite element method (FEM), to evaluate the broadband properties of dielectric materials. A characterization protocol is built to characterize dielectric materials and NN are used in order to provide the estimated permittivity. The FEM is used to create the data set required to train the NN. A method based on Bayesian regularization ensures a good generalization capability of the NN. It is shown that NN can determine the permittivity of materials with a high accuracy and that the Bayesian regularization greatly simplifies their implementation.