The goal of failures forecasting is to predict faults in the network, doing that it is possible to guarantee reliability and quality of service in real time to maintain the network availability and reliability and to initiate appropriate actions of restoration of 'normality'. The following article describes the process performed for implementing failures prediction system in LAN using artificial neural networks multilayer Perceptron. It describes the system, the tests made for the selection of the own parameters of the neural network like the training algorithm and the obtained results.