Since June 1995, in the BPI, an access to Internet has been existing. After several months, many questions are asked. Internet is here a document retrieval tool allowing an access to an infinite number of documents. But is the word document the good one for this new world, doesn't it cover several notions ? This paper tries to show the problems of identification, of reliability, of the very existence of the documents which are on the network. Are these documents « found » or « reconstructed » by the user ? Is the « in depth » research necessitated by the hypertext links a source of enrichment or a loss of information ? The Internet use as a document retrieval tool is in fact less simple than it seems and asks an important training of users and librarians.