This paper proposes an approach for the detection of a ground fault on a unit generator-transformer, based on the extraction of statistical parameters from Wavelet transform based neural network. In the simulation, the current and voltage signals were found decomposed over wavelet analysis into several approximations and details. The simulation of the unit generator-transformer was carried out using the Sim-PowerSystem Blockset of MATLAB. The statistical parameters analysis involved measurement of the dispersion and tendency factors of wavelet coefficients. Regarding of confusion matrix and receiver operating characteristic of the neural network's pattern recognition performance, the accuracy of the single line to ground fault detection of neural networks was 98.44 %. The results indicated that the tendency factor feature of wavelet transform more accurate than dispersion factor feature of wavelet transforms in distinguishing a single line to ground-fault and normal condition for a unit generator-transformer.