This study was undertaken to discover novel biomarkers for the noninvasive early diagnosis of nonalcoholic fatty liver disease (NAFLD). A methionine and choline deficient (MCD) diet was used to represent different stages of NAFLD in male C57BL/6 mice. (1)H NMR spectroscopy and principal components analysis (PCA) were used to investigate the time-related biochemical changes in mice sera induced by the MCD diet. Many serum metabolites' concentrations changed between control and MCD-fed mice. Hierarchical cluster analysis (HCA) and artificial neural networks (ANNs) were used to select the least number of metabolites to be used for the noninvasive diagnosis of various stages of NAFLD; four potential biomarkers, serum glucose, lactate, glutamate/glutamine, and taurine were selected. To verify the diagnostic accuracy of these selected metabolites, their serum concentrations were measured in healthy controls (n = 28), NAFLD patients with steatosis (n = 15), steatosis patients with necro-inflammatory disease (n = 11), and NASH patients (n = 6). On the basis of results from MCD-fed mice model, clinical tests, and previous reports, we propose using the levels of the four metabolites for diagnosing NAFLD at various stages. Furthermore, the probability of developing NAFLD at a particular stage was assessed by multinomial logistic regression (MLR) based on the clinical results of the four serum metabolites.