Schizophrenia is a chronic, severe, and disabling brain disorder. Approximate 1% of Americans have this illness. It typically occurred in young adulthood (15~45 year). Generally, schizophrenia patients usually could not identify reality and hallucination and it would make social and occupational dysfunctions. Until now, diagnosis of schizophrenia is based on observed behavior and the patient’s reported experiences (Diagnostic and Statistical Manual of Mental Disorders, DSM-V). The aim of this study is to evaluate the feasibility of classifying schizophrenia and healthy people by analyzing physiological information. In this paper, resting-state fMRI (rfMRI) data from 69 schizophrenia patients and 72 healthy people were studied. Comparing to rfMRI, patients are not required to perform any task during the rfMRI experiment so that it reduces the difficulty of data collection. We estimated the correlation matrix of 116 regions of interesting (ROI) from automated anatomical labeling (AAL). Then the resultant correlation matrix is converted to the binary graph. The network complexity analysis is applied to the binary graph to estimate the largest connected components (LCC) and the complexity of the selected distinguishable connected components to differentiate abnormal brain regions of patients from normal brains of healthy people. Finally, the extract features were fed to the linear SVM for classification. Accuracy of the proposed method can reach 71.63% through leave-one-out cross validation. Experimental results demonstrate the feasibility of schizophrenia diagnosis based on the brain network complexity of the rfMRI.