Dukes' classification is the most usually used method of staging in colorectal carcinoma. However the pitfalls of this classification may be attested by the multitude of modifications of Dukes' staging during the last sixty years. The aim of our study was to evaluate the influence of 25 variables on survival in colorectal carcinoma using the Cox model. This study was performed in 321 patients operated on for colorectal carcinoma. In the 252 patients who had had a curative resection univariate analysis showed that survival was related to 6 variables: age, juxtatumoral lymph node involvement, an abdominal mass, complicated carcinoma, duration of symptoms, and depth of tumor penetration. Stepwise analysis showed the independent influence of 7 variables on survival: age, lymph node involvement, depth of tumor penetration, duration of symptoms, direct spread, preexisting deficiencies, and sex. A byproduct of the regression coefficient of Cox' model was a numerical value for each of these predictive factors. A single composite score was obtained by adding the individual scores of factors from which a prediction of outcome can be made. Applying this model to each of the 252 patients resected for cure, we established three prognostic groups. They predict survival more accurately than the Dukes' system. Our scoring system based on multivariate analysis must successfully defined prognostic groups of patients in other series. Then, it should be recommended for staging in curative resection for colorectal carcinoma.