The mild cognitive impairment(MCI) is highly likely to develop into Alzheimer's disease, so early diagnosis of MCI is especially important. In this study, the number of global bridges were selected as the classification feature for the first time. This research used the greedy algorithm of “heap structure” to modularize the resting state functional brain network of MCI and Normal control(NC), then, the redundant functional connections in the network were removed according to the centrality of connection mediators, and the number of global bridges between modules and within modules were selected as the classification features. Support vector machine(SVM) was used as classification model to recognize NC and MCI. Results show that the average classification accuracy rate reached 92.89%, and the statistical analysis shows that there were significant differences in the number of global bridges between and within the modules between the two groups, especially in the default network and the limbic system, which was basically consistent with the previous research.