Objectives: There is an increasing interest in the neurophysiologic mechanisms of amnestic mild cognitive impairment (aMCI). However, the changes in information segregation and integration are unclear. Methods: Genuine Symbolic Nonlinear Granger Causality (GSNGC), a new coupling method, was utilized to analyze the functional connectivity differences (nodal and global levels) between aMCI and Normal Control (NC) by clustering coefficient and efficiency metrics. Finally, aMCI was diagnosed by a genetic algorithm-based support vector machine (GA-SVM) based on the GSNGC network metrics. Results: The average network connection strength in aMCI was lower than in the NC group, except for theta frequency. The significant difference of global eigenvalues (global efficiency and average clustering coefficient) in the alpha frequency band between aMCI and NC is the highest (p < 0.001). There is a significant difference in nodal eigenvalues for the temporal lobe between aMCI and NC (p < 0.05). Finally, the classification accuracy of the GA-SVM based on GSNGC network features was 93.4 %. Conclusions: The GSNGC network was an effective indicator for identifying aMCI. Loss of cognition was associated with the redistribution of the pattern of information integration. Meanwhile, network compensation mechanisms may exist in the brains of aMCI individuals.