Background and Aims: The overall survival (OS) of hepa-tocellular carcinoma (HCC) remains dismal. Bioinformatic analysis of transcriptome data could identify patients with poor OS and may facilitate clinical decision. This study aimed to develop a prognostic gene model for HCC. Meth-ods: GSE14520 was retrieved as a training set to identify differential expressed genes (DEGs) between tumor and adjacent liver tissues in HCC patients with different OS. A DEG-based prognostic model was then constructed and the TCGA-LIHC and ICGC-LIRI datasets were used to validate the model. The area under the receiver operating charac-teristic curve (AUC) and hazard ratio (HR) of the model for OS were calculated. A model-based nomogram was estab-lished and verified. Results: In the training set, differen-tial expression analysis identified 80 genes dysregulated in oxidation-reduction and metabolism regulation. After uni-variate Cox and LASSO regression, eight genes (LPCAT1, DHRS1, SORBS2, ALDH5A1, SULT1C2, SPP1, HEY1 and GOLM1) were selected to build the prognostic model. The AUC for 1-, 3- and 5-year OS were 0.779, 0.736, 0.754 in training set and 0.693, 0.689, 0.693 in the TCGA-LIHC validation set, respectively. The AUC for 1- and 3-year OS were 0.767 and 0.705 in the ICGC-LIRI validation set. Mul-tivariate analysis confirmed the model was an independent prognostic factor (training set: HR=4.422, p<0.001; TCGA-LIHC validation set: HR=2.561, p<0.001; ICGC-LIRI valida-tion set: HR=3.931, p<0.001). Furthermore, a nomogram combining the model and AJCC stage was established and validated, showing increased OS predictive efficacy com-pared with the prognostic model (p=0.035) or AJCC stage (p<0.001). Conclusions: Our eight-gene prognostic model and the related nomogram represent as reliable prognostic tools for OS prediction in HCC patients.