Stylistic analysis includes the automatic identification of the register of a document such as legal, literary... In this paper we propose a typology of criteria to distinguish between science and popular science, based on observations of Web documents, forming a part of a comparable corpus. This typology holds two dimensions: the external dimension, characterizing the web site creation context, and the internal one, characterizing the communicative tasks of the website and its documents (graphical and textual characteristics). Then the criteria of the typology will be implemented to automatically classify Web documents between scientific and popular-science register. That is a first step to the automate of a comparable corpus of Web documents.