As one of the important tools for knowledge discovery and data mining, rough set theory based on granular-ball computing has been successfully applied to label prediction and attribute reduction. However, the existing granular-ball rough set models only consider a single granulation, and cannot analyze and process data from a multi-granulation, and there are still many application scenarios that need to be considered from the perspective of multi-granulation. Based on this, this paper proposes a multi-granulation rough set based on granular-ball computing by embedding the idea of granular-ball in the multi-granulation rough set model, and discusses the relevant properties of the model. The model divides the data by setting the purity, which can effectively depict the internal relationship between the data, and thus design a position region generation algorithm for multi-granulation granular-ball rough set. Experimental analysis shows the feasibility and effectiveness of this model.