Background: Distant metastasis with an incidence of 25% in esophageal cancer(EC) represents a poor prognosis. However, there was few study for prediction of distant metastasis in EC, due to unsatisfactory specificity of clinical factors and lack of reliable biomarkers. Methods: Two hundred and ninety-nine patients were enrolled and randomly assigned to a training cohort(n = 207) and a validation cohort(n = 92). Logistic univariate and multivariate regression analyses were used to identify clinical independent predictive factors and construct a clinical nomogram. Radiomic features were extracted from contrast-enhanced CT performed before treatment, and Lasso regression was used to screen the optimal features, which were developed a radiomics signature subsequently. Four machine learning algorithms were used to establish radiomics models respectively based on the screened features. The joint nomogram incorporating radiomics signature and clinical independent predictors was developed by logical regression algorithm. All models were further validated by discrimination,caliberation, reclassification and clinical usefulness. Results: The joint nomogram had a better performance [AUC(95%CI): 0.827(0.742-0.912)] than clinical nomogram [AUC(95%CI): 0.731(0.626-0.836)]and radiomics predictive models[AUC(95%CI): 0.747(0.642-0.851),SVM algorithms]. Caliberation curve, and decision curve analysis also revealed joint nomogram outperformed the other models. Compared with the clinical nomogram, net reclassification Improvement(NRI) of the joint nomogram was improved by 0.114(0.075, 0.345),and integrated discrimination Improvement (IDI) was improved by 0.071(0.030-0.112),P= 0.001. Conclusions: We constructed and validated the first joint nomogram for distant metastasis in EC based on radiomics signature and clinical independent predictive factors, which could help clinicians to identify patients with high risk of distant metastasis.