DNA ploidy studies were performed in 188 patients operated on for rectal cancer. In order to define different risk groups of patients, a stepwise logistic regression was carried out in 138 patients who underwent abdominal "curative" resections. Thirty-seven variables were analyzed. Although several variables were significant, only three improved the prognostic value: (1) more than three positive lymph nodes (p = 0.0007); (2) macroscopic local tumor invasion (p = 0.01); and (3) DNA ploidy (p = 0.03). Standardized discriminant coefficients were used to obtain a model and format for predicting local recurrences. This is the first time that a predictive model for rectal cancer, using DNA ploidy as a variable, is reported. Based on calculated discriminant values (DV), patients can be divided into three subgroups: (1) low risk for local recurrences (DV < -1.9, n = 56)--local recurrences were observed in two patients (3.6%); (2) moderate risk (DV between -1.9 and -0.6, n = 55)--local recurrences occurred in nine patients (16.4%); and (3) high risk (DV > -0.6, n = 27)--local recurrences occurred in 14 patients (51.8%). This predictive model for local recurrences has much better prognostic value than Dukes' staging (p < 0.0001).