Summary: Concept is one of the important granularities in the process of information granulation. However, concept-based knowledge representation cannot effectively describe prototype category with family similarity. Inspired by the modern category theory, this paper proposes concept cluster to represent a group of closely related concepts, so as to depict information granules with granularities larger than concepts. And then, some important properties and calculation methods of concept clusters are studied. Finally, an information retrieval method based on concept cluster is proposed.