The aim of this study was to investigate whether it was possible to select patients with Dukes' C primary colorectal cancer who had a high risk of metastasis and recurrence. Among 434 patients who underwent surgical resection of primary colorectal cancer, 115 patients (26.5%) had Dukes' C cancer. Among them, 35 patients (30.45%) suffered from metastasis/recurrence postoperatively. Thirty-two patients with recurrence for whom complete follow-up was possible (recurrence group) and 32 patients who survived for at least 5 years without recurrence (non-recurrence group) were compared by immunohistochemical staining of resected lymph nodes for cytokeratin using two antibodies (AE1/AE3 and CAM 5.2). AE1/AE3 staining was positive in 93.85% of the recurrence group versus 68.85% of the non-recurrence group (p=0.0250), while CAM 5.2 staining was positive in 84.45% vs. 53.15%, respectively (p=0.0152). The occult neoplastic cell count (mean +/- SD) in the peripheral lymph node sinuses was 6.28+/-5.17 vs. 2.38+/-3.03 (p=0.0002) in the AE1/AE3-positive patients from the recurrence and non-recurrence groups, while it was 5.13+/-4.84 vs. 1.53+/-2.37 (p=0.0003) in the CAM 5.2-positive patients in each group. Accordingly, the immunohistochemical positivity rate for both AE1/AE3 and CAM 5.2, as well as the occult neoplastic cell count, were significantly higher in the recurrence group. These results suggest that patients with Dukes' C primary colorectal cancer who have a higher risk of recurrence can be selected by immunostaining of resected lymph nodes for cytokeratin.