The application of Radial Basis Function (RBF) neural networks for Fault Section Estimation (FSE) and fault classification and fault location within faulty section in transmission lines is presented. At the first step, a multi-way graph partitioning method based on weighted minimum degree reordering is proposed for effectively partitioning the original large-scale power system into desired number of connected sub-networks. After partitioning, the proposed scheme for each part of system consists of six RBFNNs, one network for FSE, one for fault classification and four networks for fault location one for each fault type within the faulty section. For FSE, the relay and circuit breaker states are taken as the input to the distributed FSE system, while the states (faulted or normal) of transmission lines as the outputs. For fault classification, pre-fault and post-fault samples of the three-phase currents and another input from FSE are taken as the input, while faulty phase(s) as the output. For fault location, post-fault samples of both currents and voltages of the three phases and another input from both FSE and fault classification are taken as the input, while the fault locator as the output. To validate the proposed approach simulation studies have been carried out on IEEE 11-bus system in normal and faulty conditions to train and test the RBFNN. Testing results proved that the proposed RBF networks could provide great performance for high speed relaying. It is accurate, fast and reliable.