Abstract: Objective To develop an esophageal squamous cell carcinoma-support vector machine(ESCCSVM) classifier for ESCC survival prediction, and to evaluate its performance of prognostic prediction. Methods Based on the survival data of 168 ESCC patients, we determined the correlation between ESCC survival and clinicopathological features, the expression levels of three biomarkers, 14-3-3σ, heatshock protein gp96 and macrophage migration inhibitory factor(MIF). The SVM algorithm performing in Matlab was used to develop an optimal ESCC-SVM classifier for prognostic prediction by using 128 ESCC patients randomly selected out of 168 patients as the training set and the rest 40 cases as a test set to verify the classification efficiency. Receiver operating characteristic(ROC) curve analysis was used to evaluate the potential of ESCC-SVM classifier and other prognostic factors for identification of high- and low-risk patients. Results ESCC-SVM classifier comprised sex, T stage, histological grade, lymph node metastasis, TNM stage, 14-3-3σ and gp96 as the optimal factors. ROC curve analysis showed that ESCC-SVM produced the largest AUC both in the training and validation groups (0.96, 0.86, respectively), with an accuracy of 97.7% and 90.0%, respectively, which performed significantly better than TNM system(accuracy 62.5%, 67.5%, respectively) and other clinicopathological features. SVM classifier was significantly correlated with sex, T stage, histological grade, lymph node metastasis, TNM stage and 14-3-3σ. Multivariate Cox proportional hazards regression analysis indicated that age, T stage, gp96 and ESCC-SVM were independent prognostic factors of ESCC. Conclusion The seven-feature SVM classifier for ESCC prognosis prediction would provide theoretical evidence for the prognosis, clinical treatment selection and individualized treatment.