Endoscopic ultrasonography (EUS) can differentiate between impression and submucosal tumor (SMT) but it is not known whether EUS criteria can reliably guide management.The aim of this prospective study was to assess an approach to recommend removal versus follow-up investigation based on clinical and EUS criteria, with respect to the predictive values to recognize malignancy versus benign lesions.Material and methods. Over a 7-years time period, all patients referred for the EUS assessment of submucosal upper GI lesions were prospectively enrolled. Extraluminal impressions diagnosed with EUS were not further considered. If submucosal tumors seen with EUS were clearly symptomatic or one of several parameters (tumor size >3 cm, irregular margins, inhomogeneous echotexture and/or enlarged lymph nodes) were found, resection was recommended. The remaining cases were subjected to EUS follow-up.Results. Of cases with 241 submucosal lesions, 65 had impressions and 176 had true submucosal lesions. Of the latter, 29 cases had non-neoplastic lesions (cysts, varices). In 59 cases, removal was deemed necessary due to clinical symptoms and suspicious findings in conventional endoscopy. These subjects underwent either surgical (originating layer, muscularis propria) or endoscopic resection (submucosal origin): 35.6% were malignant, more frequently in the surgical group (41.6% vs 20%). However, in 52.5% (n=31) of the 59 cases with no severe symptoms and true SMT, EUS suggested removal because of their additional criteria. Eighteen patients (12.2%) refused SMT removal and even regular EUS-based follow-up investigation. Clinical follow-up investigation by the family practitioner did not show frank malignancy in these cases (retransferal not registered). Follow-up investigation with EUS was recommended in 70 cases (mean follow-up period, 5 years; range, 1-7 years). The pattern remained unchanged in 67/70, and 2 of the 3 cases with changes underwent surgery for benign leiomyoma (patient refusal, n=1 with no change in the one-year follow-up MRI).Conclusions. An EUS strategy based on defined characteristics to remove SMT with no severe symptoms and suspicious finding in the conventional endoscopy shows a good adherence to the recommended approach and has a reasonable positive predictive value for malignancy (88%). Clinical symptoms alone or with endoscopic finding are frequently too vague to decide for a reasonable SMT resection. The chosen EUS criteria are valuable to: 1) achieve the primary resection of all potentially malignant SMT and 2) avoid to overlook them as shown by the results of the follow-up investigations with no detected malignant lesion.