OBJECTIVES To determine the efficacy with which endoscopic ultrasonography (EUS) is able to differentiate between mucosal and submucosal invasion for application of endoscopic resection. METHODS We prospectively analyzed 60 patients who were diagnosed with early cancer by conventional EUS with regard to the accuracy of mucosal neoplasia as a function of gender, age, location, size, endoscopic configuration, histological diagnosis, and method of resection. RESULTS Forty lesions interpreted as mucosal by EUS were shown histologically to include 32 lesions in the mucosa and eight in the submucosa or deeper, whereas 20 tumors interpreted as invasive cancer included six lesions in the mucosa and 14 in the submucosa or deeper. The accuracy of mucosal neoplasia (AMN) detection was 77% (true-positive and true-negative mucosal neoplasias divided by all lesions). Assessment of pure cancers without adenomatous components produced a significantly lower AMN (59%, p = 0.03) than the assessment of pure adenomas (95%) or cancers in adenomas (87%). However, there were no significant differences in the EUS assessment of intramucosal neoplasia as a function of sex, age, endoscopic configuration, size, location, or treatment. CONCLUSIONS We conclude that conventional EUS interpretation alone cannot determine the appropriate treatment for early colorectal cancer.