Abstarct: Nowadays, employing Distributed Generations (DG) in power systems is highly interested. A considerable problem of using the small generation units is Islanding phenomenon. In this project a new approach baxsed on governor signal processing is introduced for loss of main detection scenario. Recently neural network has been gained so many attentions of the researchers working on the field of system identification or modeling. In this thesis both Self-Organizing Map (SOM) and Probabilistic Neural Network (PNN) are applied to cluster and classify some signals in order to detect islanding situation. In addition, a new active scheme is introduced using the governor signal which triggers a voltage relay when islanding. Simulation results show that neural network can be confidently applied when designing a relay for islanding detection. This relay prevents any undesired consequences of islanding and reconnection of the main including out of synchronism operation of the DG.