In this paper, the authors present successful field test experience in the use of neural networks for short-term electrical load forecasting. After reviewing the importance of load forecasting as a key planning tool for a modern energy management system (EMS), they outline the advantages of using neural networks, how they are implemented, the choice of explicative variables and the selection of appropriate models. In the field test, a fully automatic load forecasting service was implemented. Numerical results are presented showing the importance of forecasted temperatures for a good load forecast and a comparison of rural and urban regions in terms of accuracy