We investigated the potential application of preoperative serum metabolomes in predicting recurrence in patients with resected pancreatic cancer. From November 2012 to June 2014, patients who underwent potentially curative pancreatectomy for pancreatic ductal adenocarcinoma were examined. Among 57 patients, 32 were men; 42 had pancreatic head cancers. The 57 patients could be clearly categorized into two main clusters using 178 preoperative serum metabolomes. Patients within cluster 2 showed earlier tumor recurrence, compared with those within cluster 1 (p = 0.034). A nomogram was developed for predicting the probability of early disease-free survival in patients with resected pancreatic cancer. Preoperative cancer antigen (CA) 19-9 levels and serum metabolomes PC.aa.C38_4, PC.ae.C42_5, and PC.ae.C38_6 were the most powerful preoperative clinical variables with which to predict 6-month and 1-year cancer recurrence-free survival after radical pancreatectomy, with a Harrell's concordance index of 0.823 (95% CI: 0.750-0.891) and integrated area under the curve of 0.816 (95% CI: 0.736-0.893). Patients with resected pancreatic cancer could be categorized according to their different metabolomes to predict early cancer recurrence. Preoperative detectable parameters, serum CA 19-9, PC.aa.C38_4, PC.ae.C42_5, and PC.ae.C38_6 were the most powerful predictors of early recurrence of pancreatic cancer.