AbstractThis paper proposes a method for daily maximum load forecasting in power systems. It is based on the integration of the regression tree and the artificial neural network. In this paper, the regression tree is used to extract knowledge or rules as a data‐mining method. That is useful for the information processing of the complicated data. As a result, the proposed method has an advantage in clarifying the cause and effect of dynamic load behavior in load forecasting. However, the regression tree does not necessarily yield good prediction results in spite of good classification. Therefore, this paper proposes a method for combining the classification results of the regression tree with the multilayer perceptron of a universal nonlinear approximator. The effectiveness of the proposed method is demonstrated in real data. © 2002 Scripta Technica, Electr Eng Jpn, 139(2): 12–22, 2002; DOI 10.1002/eej.1150