Many research efforts have been devoted to solving the problems of Boolean systems, which are currently used for Information Retrieval (IR). We propose a new model of IR, which treats the whole process of IR as a process of evidential reasoning. Our model is knowledge based, and theoretically sound. An input query provided by a user, triggers the process of evidential reasoning. The process consists of two parts: automatic query formulation and query evaluation. Automatic query formulation maps a concept given by the user into a set of textual terms. These terms, according to the pieces of evidence given by an expert, have been used by various authors to describe the concept specified in the input query. Query evaluation is an evidence-aggregation scheme, that combines all the pieces of evidence and assigns a Retrieval Status Value (RSV) to each document. A list of documents, ranked according to the RSV, is provided to the user as a response to his or her information request. In our model, inference strength between concept and subconcept is measured by conditional basic probability assignment; and this measure is discounted, chained, and combined based on the Dempster-Shafer (D-S) theory and its extension.