Automatic web page segmentation is the basis to adaptive web browsing on mobile devices. It breaks a large page into smaller blocks, in which contents with coherent semantics are keeping together. Then, various adaptations like single column and thumbnail view can be developed. However, page segmentation remains a challenging task, and its poor result directly yields a frustrating user experience. As human usually understand the web page well, in this paper, we start from Gestalt theory, a psychological theory that can explain human's visual perceptive processes. Four basic laws, proximity, similarity, closure, and simplicity, are drawn from Gestalt theory and then implemented in a program to simulate how human understand the layout of web pages. The experiments show that this method outperforms existing methods.