In this paper, we present a multilingual information retrieval system based on knowledge representation model. This system allows document indexing and information retrieval in a multilingual document collection where documents are written in different languages, though each individual document may contain text in only one language. The underlying model permits to describe the semantic of document in a multilingual context. This model, called semantic graph, is an extension of the Sowa's model of conceptual graphs where different vocabularies are available. Indeed, in this model two kinds of knowledge are identified: • Domain knowledge organises domain entity in two hierarchies of types (concept types and relation types), • Lexical knowledge associates term, belonging to a vocabulary, to concept type or relation type. Then, a same semantic graph can have different representations depending on the vocabulary chose. For example, the French representation of a semantic graph uses the French labels of types to display each graph component and the English representation of the same graph uses the English labels of types to display the graph. Our proposition has been validated in the logical information retrieval system SyDoM. The system is dedicated to the needs of virtual libraries for managing XML documents. Thanks to the semantic graph model, SyD0M develops several functionality for multilingual information retrieval. This first evaluation gives better results than traditional information retrieval system.