Objectives: To explore whether MRI and radiomic features can differentiate uterine sarcoma from atypical leiomyoma. And to compare diagnostic performance of radiomic model with radiologists. Methods: 78 patients (29 sarcomas, 49 leiomyomas) imaged with pelvic MRI prior to surgery were included in this retrospective study. Certain clinical and MRI features were evaluated for one lesion per patient. Radiological diagnosis was made based on MRI features. A radiomic model using automated texture analysis based on ADC maps was built to predict pathological results. The association between MRI features and pathological results was determined by multivariable logistic regression after controlling for other variables in univariate analyses with P < 0.05. The diagnostic efficacy of radiologists and radiomic model were compared by area under the receiver-operating characteristic curve (AUC), sensitivity, specificity and accuracy. Results: In univariate analyses, patient's age, menopausal state, intratumor hemorrhage, tumor margin and uterine endometrial cavity were associated with pathological results, P < 0.05. Patient's age, tumor margin and uterine endometrial cavity remained significant in a multivariable model, P < 0.05. Diagnosis efficacy of radiologists based on MRI reached an AUC of 0.752, sensitivity of 58.6%, specificity of 91.8%, and accuracy of 79.5%. The optimal radiomic model reached an AUC of 0.830, sensitivity of 76.0%, average specificity of 73.2%, and accuracy of 73.9%. Conclusions: Ill-defined tumor margin and interrupted uterine endometrial cavity of older women were predictors of uterine sarcoma. Radiomic analysis was feasible. Optimal radiomic model showed comparable diagnostic efficacy with experienced radiologists.