A major need of precision-oriented retrieval systems is the availability of high-level indexing languages which can allow the representation and the manipulation of elaborated concepts for both indexing and querying processes. Such elaborated knowledge representations, as the one developed in the RIME project, need in turn to be fully integrated within the underlying retrieval model to ensure proper and optimal use of this knowledge in the retrieval process. In this paper we present further developments of the fuzzy modal logic retrieval model which was first developed and experienced in the RIME project. These developments are precisely aimed towards a better integration of this model and knowledge representation. We show that Conceptual Graphs of Sowa have formal properties which allow a good formal control of this integration, and are inherently well adapted to IR requirements. This study has triggered extensions of the initial logic model and adaptations of the conceptual graphs' formalism which are also presented here. Finally the paper presents the way every notion of the theoretical retrieval model is expressed in terms of a derived operational model which has been implemented.