A new method for classifying the preclinical stages of Alzheimer's disease (AD)and positioning the related brain areas is described in this paper in order to slow the progress of AD. The method is based on multi-voxel pattern analysis (MVPA)., which is used to classify normal control (NC)participants and patients and find the changes in different brain areas with AD progression. In the classification., each voxel's blood oxygen level dependence (BOLD)signal during resting-state functional magnetic resonance imaging (rs-fMRI)was extracted as the basic features. To reduce the amount of features, principal component analysis (PCA)and two-class support vector machine (SVM)were applied to classify 62 NC participants and 162 different stages of patients, which included 47 subjective cognitive decline (SCD)patients, 60 amnestic mild cognitive impairment (aMCI)patients and 55 AD patients respectively. The accuracy of classification reached to 62.71% in SCD., 70.67% in aMCI and 86.36% in AD (all of them were classified with NC participants). Based on the accuracy, we innovatively combined 'weight vectors' in SVM with permutation test as discrimination patterns to further investigate the related brain areas. The discriminating areas, including middle cingulum (right), insula (left)., paracentral lobule (right)and middle temporal (left)are responsible for different cognitive functions and could provide a large application of AD biomarkers. Our method and results suggest the potential of real-time diagnosis and cognitive therapy because of no complex feature calculations. ? 2018 IEEE.