Background Vertebral compression fractures (VCFs) are common clinical problems that can be caused by various reasons. The differential diagnosis of benign and malignant VCFs is challengeable. This study aimed to develop and validate a radiomics model to predict benign and malignant VCFs on 18F-fluorodeoxyglucose-positron emission tomography/computed tomography (18F-FDG-PET/CT). Results Twenty-six features (9 PET features and 17 CT features) and five clinical variables (age, SUVmax, SUVpeak, SULmax, SULpeak) were selected ultimately. The area under the curve (AUC) of the radiomics and clinical-radiomics model were statistically different from the clinical model in both training group (0.986, 0.987 vs. 0.884, p < 0.05) and test group (0.962, 0.948 vs. 0.858, p < 0.05), while there was no significant difference between the radiomics model and clinical-radiomics model (p > 0.05). The accuracy of the radiomics and clinical-radiomics model was 94%, 95.0% in training group and 93.2%, 93.2% in test group. Three models all showed good calibration (Hosmer_Lemeshow_test, p > 0.05). According to the decision curve analysis (DCA), the radiomics model and clinical-radiomics model showed higher overall net benefit than the clinical model. Conclusions The PET/CT-based radiomics and clinical-radiomics model showed good performance in distinguishing between malignant and benign VCFs. The radiomics method may be valuable for treatment decision-making.