Up to 25% of all resected patients with localized colon cancer (stage UICCII) develop recurrence and die from the disease within 5 years. The aim of the study was to identify and validate a microarray-based prognosis signature for stage II colon cancer patients, using own and published data. In this study RNA tumor samples from 53 patients with stage II colon cancer were hybridized to genes chips (Affymetrix). One quarter of patients (13/53) developed a relapse (Group R), 40 patients (75%) remained relapse-free (RF group) within 5 years after surgery. For the development of a prognosis signatures different methods were used („Nearest Shrunken Centriods“ with cross-validation). In addition published prognosis signatures were applied to our data (Global Test and leave one out „cross-validation). Our gene expression data allow the identification of differentially expressed genes between the groups„R“ and „RF“. However, application of various bioinformatic methods did not generate a stable classification and prognosis signature based on our data. The application of published gene signatures (22 and 19 genes) to our data also allows a separation of the groups „R“ and „RF“ (p=0.006 or 0.014). However, the proportion of correct predictions (75.5% and 64.2%) was too low and statistically not significant. Three genes (PBK, CXC11 and CA2) mainly affected the result and showed a stronger expression in group RF. Gene expression analysis allows identification of differentially expressed genes between prognostically diverse groups in our patient cohort. However, the development of a stable prognosis signature in our data did not result in a statistically significant geneset. The application of published signatures on our data in terms of an external validation also did not reach statistical significance. These results suggest that prediction of clinical course by gene expression signatures as a basis for long-term prognosis is not or only partially possible. Nevertheless, larger studies may help to identify more stable signatures.