Colon cancer patients routinely undergo preoperative computed tomography (CT) scanning, but local staging is thought to be inaccurate. We aimed to determine if clinical outcome could be predicted from radiological features of the primary tumour. Consecutive patients at one hospital undergoing primary resection for colon cancer during 2000-2004 were included. Patients with visible metastases were excluded. Preoperative CT scans were reviewed independently by two radiologists blinded to histological stage and outcome. Images of the primary tumour were evaluated according to conventional TNM criteria and patients were stratified into 'good' or 'poor' prognosis groups. Comparison was made between prognostic group and actual clinical outcome. Hundred and twenty-six preoperative CT scans were reviewed. T-stage and nodal status was correctly predicted in only 60 and 62%, respectively. However, inter-observer agreement for prognostic group was 79% (kappa=0.59) and 3-year relapse-free survival was 71 and 43% for the CT-predicted 'good' and 'poor' groups, respectively (P<0.0066). This compared favourably with 75 vs 43% for histology-predicted prognostic groups. Computed tomography is a robust method for stratifying patients preoperatively, with similar accuracy to histopathology for predicting outcome. Recognition of poor prognosis tumours preoperatively may permit investigation into the future use of neo-adjuvant therapy in colon cancer.