Traditional risk factors cannot promote prediction capacity for the patients with coronary artery disease (CAD), who usually do not show apparent symptoms until they develop acute myocardial infarction (AMI). As such, novel predictive diagnostic strategies are essential to accurately define patients at risk of acute coronary syndrome. In this study, non-targeted metabolomic profiling using ultra-performance liquid chromatography coupled to time of flight mass spectrometry (UPLC-QTOF/MS) was performed in combination with multivariate statistical model to analyze the serum samples of patients with stable angina (n = 38), acute myocardial infarction (AMI) (n = 34) and healthy age- and gender-matched controls (n = 71). Results showed a clear distinction in metabolomic profiles between stable angina and AMI when using OPLS-DA with both positive and negative models. Internal cross-validation methods were used to confirm model validity with an area under the curve (AUROC) = 0.983. We identified various classes of altered metabolites including phospholipids, fatty acids, sphingolipids, glycerolipids and steroids. We then demonstrated the differential roles of these metabolites using multivariate statistical model. Phospholipids previously associated with CAD were shown to have lower predictive capacity to discriminate AMI patients from stable angina patients. Interestingly, ceramides, bile acid and steroids hormone such as Cer(t18:0/16:0), Cer(d18:0/12:0), dehydroepiandrosterone sulfate (VIP scores of 1.99, 1.97, 1.64, respectively), were found to be associated with the progression of CAD. These results suggest that metabolomic approaches may facilitate the development of more stringent and predictive patient criteria in the diagnosis and treatment of CAD.