BACKGROUND: Mounting evidence has shown that long noncoding RNAs (lncRNAs) can function as competing endogenous RNAs (ceRNAs) which participate in the initiation and progression of cancers. In the ceRNA network, lncRNAs, microRNAs (miRNAs) and mRNAs, communicate with and co-regulate each other. Rarely there is a systematic lncRNA-mediated ceRNA network and potential specific ceRNA pairs or triples of esophageal cancer (EC). In this study, we investigate the lncRNA-mediated ceRNA network in EC and screen the potential prognostic lncRNA biomarkers. METHODS: We obtained mRNA, miRNA, and lncRNA expression data and relevant clinical features on patients with EC from The Cancer Genome Atlas (TCGA), and used the edgR package to identify differentially expressed mRNAs, lncRNAs and miRNAs between EC samples and normal samples. The EC ceRNA network was constructed based on miRNA target prediction through the databases of miRcode, miRDB, miRTarBase and TargetScan. And then Pearsonâs correlation analysis was adopted to identify co-expression mRNA-lncRNA pairs. Finally, the robust likelihood-based survival analysis and Cox regression models were used to identify prognosis-related lncRNAs, which was evaluated by Kaplan-Meier and receiver operating characteristic (ROC) curve analysis. RESULTS: A total of 3,200 mRNAs, 131 miRNAs and 1,338 lncRNAs were identified as significantly differentially expressed in EC, of which, 30 mRNAs, 15 lncRNAs, and 8 miRNAs were incorporated in the ceRNA network. According to the ceRNA network node degrees, lncRNA MAGI2-AS3, hsa-mir-93 and TGFBR2 were the key genes. Also, the ceRNA network revealed some important ceRNA pairs and triples, such as SNX29P2-TGFBR2 and MAGI2-AS-hsa-mir-143-COL1A1. Finally, we developed a six-lncRNA signature (ZNF341-AS1, AC130324.2, AC027271.1, AL591212.1, AL732314.4 and LOC105372352), with improved diagnostic potential for EC with the area under the ROC curve of 0.93. CONCLUSIONS: our present work sheds new light on the tumorigenesis roles of lncRNA-mediated ceRNA network in EC and identifies a sixâlncRNA model that could be used as candidate prognostic signature.