In 70 patients suspected of having pancreatic cancer, we prospectively compared results of seven diagnostic tests. Subsequent exploration (of 68) and liver biopsy (of two) demonstrated pancreatic cancer in 30, pancreatitis in seven, nonpancreatic neoplasms in nine and nonpancreatic non-neoplastic disease (or no disease) in 24. For detection of pancreatic disease, the best tests were the pancreatic-function test (cholecystokinin-stimulated enzyme outputs) and ultrasonography. The pancreatic scan was nonspecific (P less than 0.001), and thermography was insensitive (P less than 0.001). Endoscopic retrograde pancreatography and arteriography were significantly more sensitive than cytologic study in diagnosis of pancreatic cancer (P less than 0.001). Therefore, when pancreatic cancer is suspected, abdominal ultrasound should be performed first, and if it is negative, a pancreatic-function test next. A positive result from either test warrants an endoscopic retrograde pancreatography for definitive diagnosis. This sequence identified 88 per cent of patients without pancreatic disease and 89 per cent with pancreatic cancer.