Web information extraction is the key part of web data integration. With the need of e-commerce website and the development of web design, web pages with multiple presentation templates arise. The current web information extraction systems are usually based on single presentation template, so web pages with multiple presentation templates can’t be extracted efficiently. This paper focuses on the extraction problem about web pages with multiple presentation templates. Four different kinds of this problem have been considered, and a novel method based on path entropy, presentation regularity and ontology knowledge is presented. The experiment indicates that this method is very promising and it achieves excellent recall and precision.