Background: Gastric schwannoma (GS) was a rare mesenchymal tumor that was difficult to distinguish from a non-metastatic gastric stromal tumor (GST). The nomogram constructed by CT features had an advantage in the differential diagnosis of gastric malignant tumors. Therefore, we conducted a retrospective analysis of their respective computed tomography (CT) features. Methods: We conducted a retrospective single-institution review of resected GS and non-metastatic GST between January 2017 and December 2020. Patients who were pathologically confirmed after surgery and underwent CT within two weeks before surgery were selected. The exclusion criteria were as follows: incomplete clinical data; CT images that were incomplete or of poor quality. A binary logistic regression model was built for analysis. Through univariate and multivariate analysis, CT image features were evaluated to determine the significant differences between GS and GST. Results: The study population comprised 203 consecutive patients (29 with GS and 174 with GST). There were significant differences in gender distribution (P=0.042) and symptoms (P=0.002). Besides, GST tended to involve the presence of necrosis (P=0.003) and lymph nodes (P=0.003). The area under the curve (AUC) value of unenhanced CT (CTU) was 0.708 [95% confidence interval (CI): 62.10-79.56%], the AUC value of venous phase CT (CTP) was 0.774 (95% CI: 69.45-85.34%), and the AUC value of venous phase enhancement (CTPU) was 0.745 (95% CI: 65.87-83.06%). CTP was the most specific feature, with a sensitivity of 83% and a specificity of 66%. The ratio of long diameter to short diameter (LD/SD) was significantly different (P=0.003). The AUC of the binary logistic regression model was 0.904. Multivariate analysis showed that necrosis and LD/SD were independent factors affecting the identification of GS and GST. Conclusions: LD/SD was a novel distinguishing feature between GS and non-metastatic GST. In conjunction with CTP, LD/SD, location, growth pattern, necrosis, and lymph node, a nomogram was constructed to predict.