This study investigated whether it is possible to detect patients who have a high risk of metastasis and recurrence after resection of stage II Dukes' B primary colorectal cancer. Among 434 patients who underwent curative resection of primary colorectal cancer, 167 (38.5%) had Dukes' B cancer. Among them, 19 patients (11.4%) suffered from postoperative metastasis or recurrence. In 17 patients with recurrence who could be followed-up completely (recurrence group) and 17 other patients who survived for at least 5 years without recurrence (non-recurrence group), immunohistochemical staining of resected lymph nodes for cytokeratin (AE1/AE3 and CAM 5.2) was performed. AE1/AE3 was positive in 76.5% and 47.1% of the patients from the recurrence and non-recurrence groups, respectively, while CAM 5.2 was positive in 52.9% and 17.6%, respectively. There were no significant differences of either AE1/AE3 or CAM 5.2 positivity between the groups. However, the occult neoplastic cell count (mean +/- SD) floating in the lymph node sinuses was significantly higher in patients from the recurrence group who were positive for AE1/AE3 or CAM 5.2 than in patients from the non-recurrence group (6.12+/-6.00 vs. 0.59+/-0.71; p=0.0019 and 3.94+/-5.06 vs. 0.29+/-0.69; p=0.0098, respectively). These results suggest that patients with Dukes' B primary colorectal cancer who have a higher risk of recurrence can be selected by immunostaining of resected lymph nodes for cytokeratin.