Identification of new prognostic factors for colon cancer with no lymph node involvement may improve the selection of patients for adjuvant chemotherapy. The aim of this study was to assess the possibility of using gene expression profiling for this purpose. Fifty patients operated on for stage II colon cancer were included. Twenty-five of these patients relapsed, while the other 25 remained disease-free for at least 5 years. MRNA was extracted from fresh-frozen biopsies and hybridized to the Affymetrix GeneChip HGU133A. One thousand six hundred random splits of the 50 patients into a training set and a validation set were considered. For each split, a prognostic combination was derived from the training set (selection of the 30 genes most differentially expressed between patients who recurred and those who did not, by linear discriminant analysis), and its prognostic performance was assessed with the validation set. On average, accuracy was 76%, sensitivity 85%, and specificity 68%. A total of 6,124 genes were included in at least one of the 1,600 predictive combinations, and 55 genes were included in more than 100 combinations. This study supports the possibility of predicting the prognosis of non-metastatic colon cancer by tumor gene expression profiling. It also shows the highly variable gene composition of predictive combinations.