There are usually some noises in the historical load data， and the accuracy of load forecasting could then be impacted. Hence， it is necessary to de-noise the noises before the load data are used for load forecasting or power system analysis. Considering both horizontal and vertical continuities of electric loads， a new method for load de-noising is presented based on a two-dimension wavelet threshold de-noising method. First， the load data is transformed into a matrix of gray-scale images by normalization. Next， the images are processed by employing the two-dimension wavelet threshold de-noising method. Finally， the de-noised data are obtained after de-normalization. The feasibility and efficiency of the developed method are demonstrated by the improvement of the load forecasting accuracy of an actual example.