Despite continuing research for precise probing and grading of prostate cancer (PC) biomarkers, the indexes lack sensitivity and specificity. To search for PC biomarkers, we used proton nuclear magnetic resonance ((1)H NMR)-derived serum metabolomics. The study comprises 102 serum samples obtained from low-grade (LG, n = 40) and high-grade (HG, n = 30) PC cases and healthy controls (HC, n = 32). (1)H NMR-derived serum data were examined using principal component analysis and orthogonal partial least-squares discriminant analysis. The strength of the model was verified by internal cross-validation using the same samples divided into 70% as training and 30% as test data sets. Receiver operating characteristic (ROC) curve examination was also achieved. Serum metabolomics reveals that four biomarkers (alanine, pyruvate, glycine, and sarcosine) were able to accurately (ROC 0.966) differentiate 90.2% of PC cases with 84.4% sensitivity and 92.9% specificity compared with HC. Similarly, three biomarkers, alanine, pyruvate, and glycine, were able to precisely (ROC 0.978) discriminate 92.9% of LG from HG PC with 92.5% sensitivity and 93.3% specificity. The robustness of these biomarkers was confirmed by prediction of the test data set with >99% diagnostic precision for PC determination. These findings demonstrate that (1)H NMR-based serum metabolomics is a promising approach for probing and grading PC.