We undertook to test the recent suggestion that measurement of immunoreactive carcinoembryonic antigen (CEA) in pancreatic secretion may be useful in diagnosis of pancreatic cancer. Using duodenal intubation and a perfusion method in 57 cases, we measured the rate of pancreatic CEA secretion into the duodenum under basal saline perfusion, alone and with continuous intravenous infusion of secretin (2 clinical units per kg per hr) and of cholecystokinin-pancreozymin (CCK, 15 Crick-Harper-Raper units per kg per hr); and we compared the CEA output with secretion of trypsin, lipase, and bicarbonate under the same conditions. Subsequent laparotomy revealed pancreatic carcinoma in 25 patients, pancreatitis in 7, other intraabdominal malignancies in 6, and benign nonpancreatic disorders in 19. CEA output rates did not differentiate all pancreatic-cancer patients from other patients in any test condition. However, pancreatic enzyme outputs were abnormal with almost 90% of cancers of the pancreatic head and with 75% of cancers of the pancreatic body and tail. For detection of pancreatic cancer, enzyme and bicarbonate outputs in response to CCK are more accurate than pancreatic CEA or bicarbonate outputs in response to secretin. Since CCK-stimulated enzyme outputs can be related accurately to malabsorption (not reported here), we prefer them to bicarbonate output for assessment of pancreatic function.