This paper introduces a novel insulation diagnosis approach for Partial Discharge (PD) pattern recognition using an ensemble Neural Network (NN) system, comprising of a limited number of NNs trained for the same purpose. The training data for the ensemble NN comprises statistical parameters obtained from different PD measurements of corona from a point-to-plane geometry. The ensemble output gives the weighted average of the output of each NN which is determined by the respective certainties of each NN output. The greatest weight is given to the output with highest certainty in its decisions. Using three NNs, it is shown that the ensemble has best accuracy of 94.8% while the Multi-Layered Perception Network (MLPN), Elman Recurrent Network (ERNN) and Radial Basis Function Network (RBFN) have independent accuracies of 94.6%, 93.6% and 83% respectively. The results show that the ensemble NN model has potential for further application to other PD scenarios.