This study evaluated the possibility of dividing patients with primary rectal carcinoma into prognostic groups before surgery based on preoperative serum levels of carcinoembryonic antigen (CEA), tissue polypeptide antigen (TPA), and an antigen defined by the monoclonal antibody C-50 (CA-50), as well as on some easily available clinical characteristics providing prognostic information. The evaluation was made both for patients who were "potentially curable" by surgery and, among those, for patients who were "potentially cured." Using the Cox regression model, the serum levels of the three tumor markers, together with the knowledge of whether or not the tumor was polypoid were combined to make up the set of variables that best predicted patient outcome. These variables and their associated regression coefficients were used to classify the patients according to prognosis. The cancer-specific mortality rate for the 24% of potentially curable patients with the best prognosis was 15%; for the 26% of potentially curable patients with the worst prognosis, the cancer-specific mortality rate was 57%. For potentially cured patients among those who were potentially curable, the cancer-specific mortality rates for patients with the best and worst prognoses were 14% and 47%, respectively. The information provided by these preoperatively available variables together was comparable with that given by Dukes' staging system, but the latter system was more informative. On the other hand, some of the preoperative variables provided information not provided by Dukes' staging system.