Web resources are more and more different, not only regarding thematic content but also related to type of document, geographic origin, level, language, etc. However, web search engines do not take into account this heterogeneity and propose only a thematic access by keywords to the documents. This paper presents a method allowing to extract homogenous corpus of web documents. This method based on link analysis uses co-citation method and focuses more specially on the notion of type of web documents.