In a 5 yr retrospective study of 70 patients, the roentgenologic and surgical assessments of carcinoma at the gastroesophageal junction were compared to the histologic findings of the resected specimens. Although the lesion was demonstrated roentgenologically in all patients, no reliable indication of its limits could be obtained, and spread was seriously underestimated by preoperative Ba studies and direct examination at operation. The lack of roentgenologic evidence may have encouraged an inadequate surgical approach and contributed to incomplete resection and the high immediate mortality. There is no satisfactory method by which spread can be defined, before or at operation, and neither preoperative Ba studies nor operative assessment should influence the surgeon to limit the resection.