The study aimed to determine the value of a novel site-specific grading system based on quantifying poorly differentiated clusters (PDC; Grade(PDC)) in colorectal cancer (CRC). A multicenter pathologic review involving 12 institutions was performed on 3243 CRC cases (stage I, 583; II, 1331; III, 1329). Cancer clusters of ≥5 cancer cells and lacking a gland-like structure (PDCs) were counted under a ×20 objective lens in a field containing the maximum clusters. Tumors with <5, 5 to 9, and ≥10 PDCs were classified as grades G1, G2, and G3, respectively. According to Grade(PDC), 1594, 1005, and 644 tumors were classified as G1, G2, and G3 and had 5-year recurrence-free survival rates of 91.6%, 75.4%, and 59.6%, respectively (P<0.0001). Multivariate analysis showed that Grade exerted an influence on prognostic outcome independently of TNM staging; approximately 20% and 46% of stage I and II patients, respectively, were selected by Grade(PDC) as a population whose survival estimate was comparable to or even worse than that of stage III patients. Grade(PDC) surpassed TNM staging in the ability to stratify patients by recurrence-free survival (Akaike information criterion, 2915.6 vs. 2994.0) and had a higher prognostic value than American Joint Committee on Cancer (AJCC) grading (Grade(AJCC)) at all stages. Regarding judgment reproducibility of grading tumors, weighted κ among the 12 institutions was 0.40 for Grade(AJCC) and 0.52 for Grade(PDC). Grade(PDC) has a robust prognostic power and promises to be of sufficient clinical value to merit implementation as a site-specific grading system in CRC.