Velocity of the acoustic signal generated by partial discharge (PD) in the transformer is considerably affected by temperature. Because of thermosiphon effect, the internal temperature of the transformer gradually decreases from the top to the bottom, which leads the acoustic signal to propagate faster in the upper part of the transformer. Besides, due to the complex internal structure of the transformer, it is difficult for the acoustic signal to propagate from the PD source to the sensor along a straight line. In order to further enhance the accuracy of PD localization results, this paper proposes a correction-iteration method with considering the influence of the insulation oil temperature rise and the internal structure of the transformer. Based on the coordinate of the PD source calculated in the last iteration, more accurate results of the next iteration can be estimated. Effectiveness of the proposed method is verified by simulations and experiments, and the obtained results indicate that the localization error is positively correlated with temperature rise. Compared with other present localization methods, such as the Newton's method, the genetic algorithm and the imperial competitive algorithm, the PD localization results estimated by the proposed correction-iteration method are more accurate.