The aim of the research conducted was the creation of a model of the phenomenon of the external friction of vegetable materials using a feed-forward artificial neural network (ANN). The network was taught using a modified error backpropagation algorithm. The best modelling accuracy was obtained for a three-layer neural network having 11 neurons in the first hidden layer, 13 neurons in the second hidden layer and 1 neuron in the output layer. The accuracy of the ANN obtained was compared to the result of a theoretical-experimental model (TEM). A multiaspect analysis of the accuracy of the models investigated was conducted. It was concluded that the model of the phenomenon of external friction which made use of the artificial neural network gave a higher accuracy of the predicted value of the friction force than did the theoretical-experimental model. The higher accuracy given by the ANN can be proved both by lower mean error values (aveBw) and smaller variation ranges (aveBw ą standard deviation Bw). In the case of the TEM, the aveBw values change from -6.9% to -5.0% and in the case of ANN from -0.8% to 1.8%. The variation ranges for all data sets are smaller for the neural network model than for the theoretical-experimental model.