Background: Endoscopic ultrasonography has been accepted as an accurate means of staging cardioesophageal cancer, but no study has focused on the variability of interobserver interpretation of images so obtained. Methods: We compared interpretations recorded by five independent observers in 46 cases examined by endoscopic ultrasonography. One observer then reviewed a subset of 28 examinations 6 months later. Results: Interpretations were in nearly full accord for uT0 and uT4 tumors (κ = ≥ 0.73), in good accord for uT1 and uT3 tumors (κ = ≥ 0.42), but in poor accord for uT2 tumors (κ = 0.16). Agreement was generally good when pertaining to invasion of adjacent organs, but no agreement was noted for invasion of the pericardium. In assessment of lymph node involvement, agreement was good for intra-abdominal, subcarinal, right lower paratracheal, and paraesophageal nodes (κ = ≥ 0.49), but poor for left lower and upper paratracheal lymph nodes. Intraobserver agreement was excellent for extent of tumor infiltration (κ = 0.91) and good for lymph nodes (κ ≥ 0.51). Conclusions: In staging cardioesophageal cancer by endosonographic ultrasonography, improvement is needed in cases of uT2 tumors and of tumors invading the pericardium, and in assessment of lymph nodes in the upper mediastinum. (Gastrointest Endosc 1996;43:20-4.)