Abstract: Rough set theory is a useful method which can effectively deal with imprecise, uncertain information in the information system. Granular computing is a new field of artificial intelligence, multiple granulation is a core concept of granular computing. Multi-granulation rough set is a new research direction of rough set theory, which combined with rough set theory and the idea of granular computing. In the view of granular computing, an equivalence relation is a granulation which composed of several attributes, and a partition of the universe based on a equivalence relation can be regarded as a granularity space. Hence in the multi-granulation rough set, based on different equivalence relations, the universe can be divided into several granularity spaces and the approximation of target concept can be carried out from the multiple granularity spaces. Granularity reduction is one of the important tasks of the multi-granulation rough set research. It is the deletion of unnecessary granularity under the premise of no affection to the target concept or decision rules. Information quantity is introduced into the lower-approximate distribution reduction of pessimistic multi-granulation rough set and the information quantity of a granularity has been defined in the lower-approximate distribution reduction of pessimistic multi-granulation rough set. Then, based on the information quantity, the importance of a granularity has also been defined. A heuristic granularity reduction algorithm of pessimistic multi-granulation rough set is presented. The experimental results show the validity of the algorithm, which provide a theoretical basis for the granularity reduction of multi-granulation rough sets