Dukes' stage is the most powerful indicator of patient outcome for colorectal cancer. Several cancer survival studies have considered other prognostic variables, but results are often conflicting. We sought to assess the independent value of several clinical and morphological variables in defining colorectal cancer specific survival. 397 colorectal cancer patients diagnosed from 1984 to 1986, and registered in a large bowel cancer registry instituted in a local health district of Northern Italy, were actively followed-up until 31 December 1991. Univariate and multivariate survival analyses were carried out in colon and rectal cancer cases, separately, using the actuarial life-table method and Cox proportional hazard regressions. Crude and specific 5-year survival rates were 37.5 and 41.4%. In univariate analysis, TNM (tumour, nodes and metastases) stage was the strongest predictor of prognosis in both sites. Other variables significantly related to survival were age of patient at diagnosis and pattern of tumour growth in colon cancer, type of differentiation and pattern of tumour growth in rectal cancer. In multivariate analyses, after adjusting for stage, age had a weak but significant negative effect on colon cancer survival, whereas rectal tumours with the infiltrating type of growth had a significantly worse prognosis than those with the expanding type. Colorectal cancer survival should be analysed in the main large bowel subsites in order to define high-risk groups within each TNM stage category.