Abstract— This paper presents a digital protection technique based on combined Discrete Fourier Transform (DFT) and artificial neural network (ANN) for discrimination between the magnetizing-inrush and internal-fault currents in three-phase power transformers. A full-cycle DFT is firstly applied as a preprocessing module to extract distinctive features, namely the magnitudes of the fundamental and second harmonic frequency components I1 and I2, respectively, from transient differential phase currents. The 3-phase current signals are sampled at a sampling rate of 20 samples per cycle. The features of phases a, b and c are then used to calculate the second harmonic ratio (SHR), I2/I1. Secondly, the three SHRs are fed into an ANN for classifying the transient phenomenon into either magnetizing inrush or internal-fault current. The task of the ANN unit is to develop a block signal when the SHR exceeds a threshold value. As a result, a needless relay tripping when a transformer has an inrush current can be avoided. The ANN has the architecture of an input layer, hidden layer, and output layer. The input layer has three neurons representing the SHR of each differential phase current. The neurons of the hidden layer were selected based on speed and accuracy. The output layer has one neuron with an output 0 (no trip) for inrush current or 1 (trip) for internal fault. The ANN has been trained using Levenberg-Marquardt (LM) algorithm with log-sigmoid transfer functions in the hidden and output layers, respectively. Training and testing patterns of inrush and fault currents over a wide range of inception angles have been obtained by computer simulation of a 3-phase non-linear transformer bank using MATLAB/Simulink. Simulation results show that the proposed technique can be considered as an effective digital protection approach for fast and accurate discrimination between inrush and internal fault-currents of power transformers.