In order to improve the early diagnosis of MCI, the directed functional brain network based on granger causality analysis (GCA) was constructed and used to identify subjects across normal subjects for control (NC) , patients with early mild cognitive impairment (EMCI) and those with late mild cognitive impairment (LMCI) , by examining resting-state functional magnetic resonance imaging (rs-fMRI) data.During the section of feature selection, the significantly different measures of local and global graph attributes, among the three groups, were selected as classification features by applying two-sample t-test.Subsequnetly, support vector machine (SVM) was used to classify the data, and one-way ANOVA was used to find out brain regions with significant difference on the designated network features between every pair of the three groups.The results show that the achieved accuracy was satisfactory for classification in the study.Moreover, some brain regions showed significant difference among the three groups, being consistent with previous findings to large extent.