In a prospective study from 1988 to 1990, 132 patients with suspected pancreatic tumor were examined with endoscopic ultrasound (EUS), transabdominal ultrasound (US), computed tomography (CT), and ERCP. The final diagnosis of 102 pancreatic tumors of different origin (76 malignant and 26 inflammatory tumors) and the exclusion of a pancreatic tumor in 30 patients was made by operation (N = 47), puncture (N = 36), autopsy (N = 3), or follow-up of a mean of 51 weeks (N = 46). Sensitivity and specificity in pancreatic tumor diagnosis were significantly higher for EUS (99% and 100%) than for US (67%/40%) and CT (77%/53%) and equal to ERCP (sensitivity 90%). This was even more obvious in small pancreatic tumors of 3 cm and less. However, as with the other imaging procedures, EUS was not able to differentiate reliably malignant from inflammatory pancreatic masses (accuracy 76% for malignancy and 46% for focal inflammation). From analysis of the endosonographic pattern of pancreatic tumors, no consistent morphologic features were identified which could have been specifically attributed to malignant or inflammatory masses. Our results show that EUS is superior to US and CT and equal to ERCP in pancreatic tumor diagnosis. In contrast to the indirect evidence obtained by ERCP, EUS provides direct visualization of tumor size and shape in almost all patients examined. Thus, EUS should be considered early in the evaluation of patients with suspected pancreatic tumors.