Hepatocellular carcinoma (HCC) is the leading cause of cancer-related death worldwide, because recurrence often occurs in most HCC patients undergoing hepatectomy. It is necessary to identify patients with high risk for recurrence and adopt effective therapies. An obstacle to monitor patients at high risk for poor prognosis has been the lack of useful predictive biomarkers. Fortunately, recent progress in system biology allows to screen the biomarkers for HCC prognosis in a high-throughput manner. In this study, we performed systematic Kaplan-Meier survival analysis of the whole mRNA transcriptomics based on the Cancer Genome Atlas project (TCGA) and developed a three-gene prognostic signature composing of three genes UPB1, SOCS2 and RTN3. The model was validated in two independent microarray data sets retrieved from Gene Expression Omnibus (GEO) and the expression pattern of these three predictive genes in HCC was confirmed by western blot and immunohistochemistry with our HCC samples. In conclusion, our results showed that this three-gene signature has prognostic value for HCC patients.