In order to save space of newspapers and magazines in an electric way

it is important to classify their contents automatically after they are scanned into computer. An algorithm which can establish optimal trees for the description of document structures in document segmentation is given in this paper. We can get better understanding of the document structures by using this method.