Prognosis of colorectal cancer patients that show similar histopathology may vary substantially. An attempt was made to improve prognosis by the self‐learning classification program CLASSIF1, based on automated multiparameter analysis of quantitative histochemical and clinical parameters of 64 colorectal carcinomas and adjacent normal mucosae. The histochemical parameters applied were the oxygen‐insensitivity assay of glucose‐6‐phosphate dehydrogenase (G6PDH) activity, a valid discriminator between normal and cancerous mucosae, and related parameters CuZn‐ and Mn‐superoxide dismutase (SOD) levels, and lipid peroxidation (LPO) capacity. Data were processed on the basis of a postoperative follow‐up of minimally 32 and maximally 56 months. CLASSIF1 selected the parameters oxygen insensitivity of G6PDH activity, CuZn‐SOD and Mn‐SOD levels, LPO capacity, lymph node metastasis, Dukes' stage, and age for the highest prognostic value. On the basis of these selected parameters, CLASSIF1 correctly predicted favorable outcome in 100% of the surviving patients and fatal outcome in 64% of the deceased patients. G6PDH activity appeared to be the major information carrier for CLASSIF1. On the basis of G6PDH activity parameters alone, 96% of the surviving patients and 55% of the deceased patients were correctly classified. In comparison, estimation of prognosis on the basis of Dukes' stage alone resulted in 71% correctly classified surviving patients and 61% of patients who died. It is concluded that the self‐learning classification program CLASSIF1, on the basis of quantitative histochemical and clinical parameters, is the best prognostic estimator for colon cancer patients yet available. Cytometry (Comm. Clin. Cytometry) 38:176–183, 1999. © 1999 Wiley‐Liss, Inc.