Pancreatic cancer is a malignant tumor with the worst prognosis among all cancers. At the time of diagnosis, surgical cure is no longer a feasible option for most patients, thus early detection of pancreatic cancer is crucial for its treatment. Metabolomics is a powerful new analytical approach to detect the metabolome of cells, tissue, or biofluids. Here, we report the application of H-1 nuclear magnetic resonance (NMR) combined with principal components analysis to discriminate pancreatic cancer patients from healthy controls based on metabolomic profiling of the serum. The metabolic analysis revealed significant lower of 3-hydroxybutyrate, 3-hydroxyisovalerate, lactate, and trimethylamine-N-oxide as well as significant higher level of isoleucine, triglyceride, leucine, and creatinine in the serum from pancreatic cancer patients compared to that of healthy controls. Our data demonstrate that the subtle differences in metabolite profiles in serum of pancreatic cancer patients and that of healthy subjects as a result of physiological and pathological variations could be identified by NMR-based metabolomics and exploited as metabolic markers for the early detection of pancreatic cancer.