In this paper, an electrical method is used to detect and locate partial discharge (PD) inside the winding of a three phases, 45 kVA transformer. PD is created in a separate cell outside the transformer winding using a needle-plane electrode arrangement and injected to each of the four taps in each phase of the transformer winding. The PD signal was measured as a voltage drop across a resistance connected to the ground. Different forms of PD are generated inside the transformer, i.e. sharp edge, bubble and surface discharge. Seventy PD signals are measured at each location for each PD source. Statistical features (variance, skewness and kurtosis) extracted from the measured signals have been used as an input feature to a feed-forward back-propagation artificial neural network. High recognition rate for both PD source and location has been achieved.