In this thesis an attempt to derive word classes from word-endings using a neural network is done. The network which uses the back-propagation algorithm was trained with pairs of word-endings and word categories. A brief summation of some important concepts in the field of abstract neural networks is given. Following this the actual task is attacked and its results discussed. The results (the net gives a better result than a pure guess) are indicating that to fully achieve the goal improvements are necessary. Suggestions for these improvements and future works are concluding this paper.