'Tumour response', defined as clinical or radiological evidence of tumour shrinkage is frequently regarded as an objective of chemotherapy, rather than as a predictor of prolonged survival. This study has assessed whether a fall in the serum CEA concentration after chemotherapy for disseminated colorectal cancer is a predictor of prolonged survival and compared it with tumour response as a predictor of survival. There was a 37% improvement in median survival among patients whose serum CEA concentration fell after chemotherapy (70% of patients treated) compared with patients whose serum CEA did not fall. The use of greater than 25% clinical or radiological tumour shrinkage as a predictor of prolonged survival identified a smaller proportion (36%) of patients in whom there was a 52% prolongation in median survival compared with patients whose tumours shrank less than 25%, or did not shrink. Proportional hazards regression analysis suggested that tumour shrinkage was a stronger predictor of survival. A fall in serum CEA concentration, however, identified a group of patients whose tumours did not shrink, but who had a 27% improvement in median survival compared with those whose tumours did not shrink and whose serum CEA concentration did not fall. Monitoring of the serum CEA during the first two months of treatment appears to provide a sensitive and economical means of identifying those patients whose survival is likely to be prolonged by chemotherapy for colorectal cancer.