The interconnection of large amounts of nontraditional generation causes problems in a network designed for 'conventional' operation. A microgrid consists of a combination of generation sources, loads and energy storage interfaced through fast acting power electronics. The aim of operating microgrid sub-systems is to move away from considering DG as badly behaved system components, of which a limited amount can be tolerated in an area, to 'good citizens'. The paper is focused on the development of a novel energy management system (EMS) based on the application of neural networks (NN). The EMS is able to autonomously make decisions and determine hour by hour the correct dispatch of generators with the final goal of minimising the global energy costs. The application of the EMS to a small test case has showed significant saves in the global energy bill by using typical Italian energy and fuel price data