This study investigated whether the Japanese radical lymph node dissection (J-LND) method was useful for improving the survival and outcome in patients undergoing surgical resection of primary colorectal cancer. The subjects were 434 patients with primary colorectal cancer treated over 17 years. The 10-year survival (10-YS), the number of retrieved and metastatic lymph nodes (LN), the extent of lymph node dissection (D0-D3), and the extent of lymph node metastasis (n0-n4) were compared with Dukes' classification by the Kaplan-Meier curves, log-rank test and multivariate analysis. Patients with a D number larger than their n number (D>n group) were defined as being treated according to J-LND principles, while those with a D number equal to their n number were used as controls (D=n group). Among Dukes' B patients, there was a significant difference of 10-YS between those with retrieval of > or =17 LN or < or =16 LN (p=0.0106). Among Dukes' C patients, a significant difference of 10-YS was observed between those with 1 metastatic node or > or =3 metastatic LN (p=0.0401). A significant difference of 10-YS was also noted between Dukes' C patients with D>n or D=n (p=0.0282). Multivariate analysis showed that retrieval of < or =16 LN (HR=9.051) and intramural invasion (se,si/a2,ai; HR=6.313) were independent determinants of 10-YS in Dukes' B patients, while D=n (HR=2.354) and > or =3 metastatic LN (HR=2.210) were independent determinants in Dukes' C patients. These results suggest that the J-LND method should be performed to retrieve at least 17 nodes when serosal dimpling of the primary tumor is observed during surgery. Effective post-operative adjuvant therapy, such as combination chemotherapy and/or radiotherapy, should be provided for Dukes' C patients with D=n and/or > or =3 metastatic nodes on histopathological examination.