Purpose: Models for predicting perforation during endoscopic resection (ER) of gastric submucosal tumors (SMTs) originating from the muscularis propria (MP) are rare. Therefore, this study was conducted to determine important parameters in endoscopic ultrasonography (EUS) images to predict perforation and to build predictive models. Methods: Consecutive patients with gastric SMTs originating from the MP who received ER from May 1, 2013 to January 15, 2021 were retrospectively reviewed. They were classified into case and control groups based on the presence of perforation. Logistic multivariate analysis was used to identify potential variables and build predictive models (models 1 and 2: with and without information on tumor pathology, respectively). Results: In total, 199 EUS procedures (194 patients) were finally chosen, with 99 procedures in the case group and 100 in the control group. The ratio of the inner distance to the outer distance (I/O ratio) was significantly larger in the case group than in the control group (median ratio, 2.20 vs. 1.53; P<0.001). Multivariate analysis showed that age (odds ratio [OR], 1.036 in model 1; OR, 1.046 in model 2), the I/O ratio (OR, 2.731 in model 1; OR, 2.372 in model 2), and the pathology of the tumors (OR, 10.977 for gastrointestinal stromal tumors; OR, 15.051 for others in model 1) were risk factors for perforation. The two models to predict perforation had areas under the curve of 0.836 (model 1) and 0.755 (model 2). Conclusion EUS was useful in predicting perforation in ER for gastric SMTs originating from the MP. Two predictive models were developed.