Granular computing is an approach for knowledge representing and data mining. With the view point of granular computing, the notion of a granule is interpreted as one of the numerous small particles forming a larger unit. In many real-life applications, there are different granules at different levels of scale in data sets having hierarchical scale structures. Many people apply granular computing for problem solving by considering multiple levels of granularity. This allows us to focus on solving a problem at the most appropriate level of granularity by ignoring unimportant and irrelevant details. In this paper, rule extraction based on the local optimal granularities in incomplete multi-granular decision systems is explored. Firstly, the concept of incomplete multi-granular decision systems is introduced. Then the notions of the optimal granularity and the local optimal granularity in consistent incomplete multi-granular decision system are defined, and the approaches to attribute reduction and rule extraction based on the local optimal granularities are illustrated; the generalized decisions in inconsistent incomplete multi-granular decision systems are further introduced, the generalized optimal granularity and the generalized local optimal granularity are also defined, and the approaches to attribute reduction and rule extraction based on the generalized local optimal granularities are investigated. Finally, the experimental results on the public datasets are discussed.