Among 371 patients with primary colorectal cancer, 54 patients suffered from recurrence/metastasis (recurrence group) and 317 survived without recurrence for at least 5 years (non-recurrence group). The clinicopathological characteristics of the 2 groups were compared and occult neoplastic cells (ONCs) in the lymph node sinuses were detected by cytokeratin immunohistochemistry. There were significant differences of the following factors: venous invasion (v-) vs. (v+) for Dukes' A patients (p=0.0315); harvested lymph nodes (LN) <or=14 vs. >or=15 for Dukes' B patients (p=0.0388); (v-) vs. (v+) (p=0.0059), lymphatic invasion (ly-) vs. (ly+) (p=0.0435) for Dukes' A and B patients combined; D>n vs. D=n (p=0.0033), depth of tumor invasion <or=ss/a1 vs. >or=se/a2 (p=0.0329) for Dukes' C patients. When the detection of >or=3 ONCs was defined as positive, the sensitivity, specificity, PPV, and NPV were respectively 77%, 100%, 100% and 71% in Dukes' B patients, as well as 75%, 72%, 73% and 74% in Dukes' C patients. The high-risk groups for recurrence/metastasis were identified by the following criteria: (v+) and (ly+), <or=14 LN, and ONCs (+) of those with all factors for Dukes' A and B patients (selection rate, approximately 23.5%), while the criteria were D=n, >or=se/a2, and ONCs (+) of those with >or=2 factors for Dukes' C patients (selection rate; approximately 21.2-37.5%). These factors seem to be appropriate for separating patients into high-risk and low-risk groups of colorectal cancer recurrence/metastasis.