To acquire the rules of inconsistent decision table, the inconsistent decision table is granulated into different granular spaces from fine to coarse in the perspective of attribute multi-granulation. By defining the consistent granular relation matrix and inconsistent granular relation matrix, as well as mining the heuristic information hidden in the matrices, the certain rules and uncertain rules in different granular space are acquired. Finally, the proposed algorithm is illustrated by an example and verified by UCI test set. The experiment comparison with existing algorithms is done to show the feasibility and effectiveness of the proposed algorithm, which also proves that the acquired rules have better generalized ability.