Only 60% of patients having radical surgery for rectal cancer are cured of their disease. The ideal system of classification would identify just two categories--the cured and those who will die of their disease. Specimens from 379 patients who had undergone radical surgery for rectal cancer more than 20 years ago were re-examined in order to identify discrete pathological variables that independently influence long-term survival. The selected variables were given weighted scores and the score range was divided to provide four prognostic groups. The model was tested on a second data set comprising 331 patients and gave similar results. The new prognostic classification is simple to use and is superior to staging by the method of Dukes because it places twice as many patients into groups that provide a confident prediction of clinical outcome.