The power transformer protection plays vital role in power systems. Any power transformer protective scheme has to take into account the effect of magnetising inrush currents. Since during the energization of the transformer, sometimes results in 10 times full load currents and can cause mal operation of the relays. The ratio of the second harmonic to the fundamental harmonic of the inrush current is greater than that of the fault current. To avoid this we go for conventional protection scheme by sensing the large second harmonic. The second harmonic in these situations might be greater than the second harmonic in inrush currents. The differential power method has the disadvantage that the need to use voltage transforms and increased protection algorithm calculation cost. Neural networks have the disadvantage that it requires a large of learning patterns produced by simulation of various cases. This paper describes the discrimination between internal faults and inrush currents in power transformers using the wavelet transform based feature extraction technique. It is shown that the features extracted by the wavelet transform have a more distinctive property than those extracted by the fast Fourier transform due to the good time and frequency localization characteristics of the wavelet transform. The performance of this is demonstrated by simulation of different faults and switching conditions on a power transformer using MATLAB software.