Objective：This study aims to identify novel ESCC prognosis related biomarkers and construct novel prognostic predictive model for ESCC. Methods：Candidate gene strategy was adopted to analyse the expressions of key members of matrix metalloproteinases：MMP1，MMP2，MMP3，MMP7，MMP8，MMP9 and MMP10 in ESCC tissues. Immunohistochemistry was applied to detect MMP3 protein expression in 315 ESCC（Training：197 cases，Validation：118 samples）tissues. Cox proportional hazard regression model was used to evaluate the association between MMP3 expression and ESCC prognosis. Finally，we constructed ESCC prognosis predictive model by integrating MMP3 and clinicopathologic factors. Results：MMP1，MMP3 and MMP10 showed aberrant increased expression in ESCC tumor tissues. Expression of MMP3 protein in tumor tissues was significantly upregulated than that in adjacent tissues. In training dataset，the overall survival（OS）of MMP3 positive patients（22.22%）was significantly lower than patients with negative expression of MMP3（49.25%，HR=2.09，95% CI：1.45-3.03，P < 0.001）. Consistent results were further observed in validation and combined datasets. In addition，the predictive ability of model that integrated MMP3 and clinicopathologic factors was higher than that contained the TNM stage alone（AUC：0.733 vs. 0.689）. Conclusion：MMP3 acted as an ESCC prognosis related biomarker，predictive model that integrated MMP3 and clinicopathologic factors could predict the prognosis of ESCC patients accurately.