Objectives: Serum carcinoembryonic antigen (CEA) has been widely used for postoperative surveillance for colorectal cancer. However, serum CEA has a poor diagnostic accuracy for detecting recurrence. We tested the hypothesis that determining cutoff values according to the preoperative serum CEA levels would enhance the diagnostic accuracy. Methods: Serum CEA was measured before and 1-6 months after surgery in 783 patients with curatively resected colon cancer from 2005 through 2013. The cutoff values during surveillance were determined separately according to preoperative serum CEA levels. Results: In patients with negative preoperative serum CEA, the diagnostic accuracy for recurrence was 89.1% when a postoperative cutoff value was set at 5 ng/mL. However, in patients with positive preoperative serum CEA, the diagnostic accuracy was 58.4% when a postoperative cutoff value was set at 5 ng/mL, and was 75.6% when a cutoff value was set at 8 ng/mL. Among patients with positive preoperative serum CEA, the recurrence-free survival rate was significantly lower in patients with a serum CEA of ≥8 ng/mL than those with a serum CEA of <8 ng/mL (p = 0.0018). Conclusions: The diagnostic accuracy of serum CEA for recurrence is enhanced by separately setting cutoff values according to preoperative serum CEA.