In order to deal with the heterogeneous data including categorical attributes and numerical attributes from the multi-granularity and multi-level perspective effectively, double granulate criterion is built on different attribute sets sequence and different neighborhood radii. The neighborhood multi-granulation rough set model based on the double granulate criterion is proposed. Several relevant properties of the model are given, and an attribute reduction algorithm of the proposed model is presented. The reduction result can choose the appropriate attribute set and neighborhood radius flexibly according to the need of practical problems. The effectiveness of the proposed model and algorithm is verified by some examples.