Seventy-six cancers with involved lymph nodes but with limitation of direct spread in continuity to the bowel wall (Astler-Coller C1 cases) were matched with Astler-Coller C2 cases for clinical variables, macroscopic appearance of tumor, grade of differentiation, and number of positive lymph nodes. Despite this stringent matching, spread was shown to be an important prognostic variable in univariate survival analysis. Estimated five-year survival for Astler-Coller C1 cases was just below 80 percent, equivalent to B2 (Dukes' B) cases. When spread was analyzed in the presence of additional prognostic variables (character of invasive margin and lymphocytic infiltration) by multivariate modeling, its independent prognostic status was maintained. Improved survival for C1 cases was not explained by a lower incidence of local pelvic recurrence. Mechanisms to account for the better prognosis are proposed. This study reaffirms the importance of multivariate techniques of analysis in the assessment of prognosis of patients with rectal cancer.