Modifications of Dukes' (1932) classification of rectal tumours have led to confusion. From the data of 2518 patients who had undergone curative colorectal surgery the interrelationships between tumour penetration, grade, vascular invasion and pattern of lymph node involvement have been examined and their individual relevance to survival determined. Subdivision of Dukes' A cases into those confined to the muscularis mucosae (A) and those penetrating into, but not through, the bowel wall (B1) should be abandoned. Despite interrelationships between lymph node status, grade of tumour and vascular invasion, they all contribute prognostic information independent of each other. Apical lymph node involvement, more than four lymph nodes involved and extensive primary tumours with nodal involvement all carry a bad prognosis. Although interrelated each variable is individually relevant. However, subgroups of patients with Dukes' C tumours have an observed survival significantly better than expected. When few lymph nodes are involved or the primary tumour is confined to the bowel wall but lymph nodes are involved, the expectation of life is equivalent to Dukes' B.