Pulmonary embolism (PE) is a life-threatening disease. Our aim was to search for potential biomarkers using modern high-throughput metabolomics methods to improve diagnostic efficacy. The discovery cohort included 60 participants, including 30 PE patients and 30 healthy individuals. The validation cohort had 40 participants, including 20 PE patients and 20 healthy individuals. Gas chromatography-mass spectrometry (GC-MS) was combined with multivariate data analysis to determine serum metabolic profiles in patients with PE and healthy controls. The distribution of metabolic profiles in the two cohorts was assessed by unsupervised principal component analysis (PCA) and supervised partial least squares discriminant analysis (PLS-DA). Sixteen metabolites were selected from the ranked variable of predictive importance (VIP) scores and applied to the correlation analysis of PE-related clinical indicators. Four metabolites that were correlated with D-dimer levels were selected, including L-tryptophan, N-α-acetyl-L-lysine, dopamine, and N2-acetylornithine. Finally, the AUC values were calculated to be 0.958 (95% CI: 0.9122-1) for the combined biomarker panel including the 4 specific metabolites in the discovery cohort and 0.963 (95% CI: 0.9122-1) in the validation cohort. The results suggest that these four specific metabolites can be used as diagnostic biomarkers to improve diagnostic efficacy in pulmonary embolism.