We compared a semi empirical and an empirical model. The empirical model is a multilayer ANN. The semi empirical model is a custom multilayer ANN. It is a structured model, and we define the structure by hand before the training of the network. The influence of the neuron numbers on the accuracy of the models was also investigated by statistical approach. We found that the custom multilayer ANNs developed like this, are suitable for modelling the drying process of agricultural materials. They also provide the ability to improve the applicability of the empirical models. Furthermore, the semi empirical model has a higher sensitivity on neuron number applied. Keywords: drying; energetics; artificial neural network; semi-empirical model.