Artificial Neural Networks (ANNs) have had a great impact on solving a large amount of different problems. However, their development is usually a slow process in wich the human expert has to test several architectures until finding the one that achieves the best results for the solution of a certain problem. To overcome this problem, a new technique for automatically developing ANNs is used in this work. This technique uses Genetic Programmning (GP) to obtain ANNs. Moreover, the networks returned by GP have been simplified, and therefore they have a small amount of neurons for solving the problem. The results obtained have been compared with other ANN generation and training methods with Evolutionary Computation (EC) techniques in order to measure the performance of the system. To do this, some of the most used test databases were used to perform several test and compare the results with other tools. The results of these comparisons showed that the system achieved good results comparable with already existing techniques and, in most of the cases, they worked better than those techniques. This techniques has an additional advantage: after a short analysis of the network, it is possible to discriminate the variables needed for obtaining the results, wich gives new knowledge from the problem.