The application of neural networks to electrical power systems has been widely studied by several researchers [1-7]. Nevertheless, almost all the studies made so far have used the structure of neural network of back-propagation with supervised learning. In the present paper some of the more recent models particularly those that use combined non-supervised/supervised learning applied to the classification of faults in transmission lines are analyzed. In this work the following models are considered: (i) back propagation network (BP); (ii) feature mapping network (FM); (Hi) radial base function network and (iv) learning vector quantization network (LVQ). Special emphasis is made in the performance comparison in terms of the size of the neural network, the learning process, the classification precision and the robustness for generalization. The result of this work provides guides on how to select a neural network from a diversity of possibilities of neural network architecture for a specific application [7].