Purpose: Classification based on a combination of molecular and pathologic predictors had never been done using hierarchical cluster analysis. For this purpose, we identified prognostic classification based on molecular predictors, pathologic and molecular predictors, and compared their respective prognostic efficacy together with that of tumor-node-metastasis (TNM) stage. Moreover, we investigated the prognostic significance of molecular classification in different TNM stage. Experimental Design: Six pathologic predictors (p) and 13 immunohistochemical predictors (m) were investigated in 221colorectal carcinomas. Unsupervised hierarchical clustering analysis was done to group the data. Survival analysis was done by Kaplan-Meier method and log-rank test, and by multivariate COX proportional hazard model. Results: Six pathologic predictors and four molecular predictors were of significant prognostic value (P V 0.05). One molecular predictor showed a trend toward significance (P = 0.085). Hierarchical clustering analysis was done based on different combinations (5p, 13m, 5m, 5p13m, and 5p5m), and distinct groups were produced except 5p (the TNM stage was excluded). Groups identified by 5m (P = 0.053) and 5p5m (P = 0.000) showed significant differences in prognosis. Groups identified by 5p5m andTNM stage were confirmed as the independent prognostic factors in a multivariate COX proportional hazard model. Moreover, groups identified by 5m could predict different prognoses in patients with stage II disease. Conclusions: Classification based on pathologic and immunohistochemical predictors is superior to that based only on molecular predictors on prognosis. Classification based on 5m could identify additional different prognoses in patients with stage II disease.Colorectal carcinoma is one of the leading causes of cancer mortality worldwide. Accurate prognosis analysis will greatly facilitate clinical decision of the best treatment plan and reduce the healthcare costs. Up to now, tumor-node-metastasis (TNM) staging has remained the most widely used system, but the patients operated on at the same TNM stage do not necessarily have the same prognosis. Recently, with the identification of numerous molecular predictors, a more accurate staging system is expected. There have been endeavors to build a prognostic evaluation system based on molecular markers. For example, Lyall et al. identified a prognostic immunohistochemical marker profile in 90 stage III colorectal carcinomas by unsupervised hierarchical cluster analysis of 23 markers (1). Knosel et al. evaluated 11 immunohistochemical markers in 270 colorectal carcinomas (2). However, because the pathologic and molecular predictors were always classified separately in the past (1 -5), it is still unknown whether the molecular staging system would yield more accurate prognosis analyses than the traditional TNM staging system.Unsupervised hierarchical clustering analysis is a common method to profile the gene expression or tissue microarray data. For example, it h...