In this paper we present an approach for the query refinement called Librarian Agent Query Refinement Process, which simulates the role a human librarian plays in the searching for information resources in a library. The process consists of two phases: Problem Discovery, in which the potential problems (ambiguities) of the initial query are discovered and estimated and Query Refinement, which analyses these ambiguities in order to provide suitable modification of that query that increases the precision and recall of the searching process. The Agent performs comprehensive analyses of a query, based on the structure of the used vocabulary in an information portal and the content of its information repository. Moreover, the information about the previous users? behaviour and the activities of the current user are used for making the query refinement process more efficient (i.e. usage-driven and personalized). From the user?s point of view, the main advantage of such an approach is that it enables a user to satisfy his information need in a more intuitive and efficient way.