OBJECTIVE To analyze the non-invasive indexes for predicting esophageal varices (EV) in liver cirrhosis, and to establish a model for predicting the degree of EV. METHODS A total of 294 patients with liver cirrhosis and portal hypertension were divided into the following groups according to EV grade as assessed by endoscopy: non-EV and grade I EV, grade II EV and grade III EV. The non-invasive EV predictive measures of liver stiffness (LS), platelet (PLT) count, spleen thickness (ST), PLT/ST ratio, portal vein diameter, portal vein flow velocity and Child-Pugh score (CPS) were assessed by univariate analysis and multivariate logistic regression analysis, and used to generate a predictive model. The t-test, chi-square test, logistic analysis and receiver operating characteristic (ROC) curve were used in statistical analyses. RESULTS The area under the ROC for the new model was 0.990. The best cutoff value for the score was 0.898, as defined from the ROC. The sensitivity of the model was 96.5%, and the specificity was 99.2%. CONCLUSIONS The model for predicting EV was composed of LS, PLT count, ST, PLT/ST and CPS, which was accurate and sensitive, and could be used to predict EV in clinic.