Background Early diagnosis and effective treatment of liver hepatocellular carcinoma (LIHC) are keys to improving the prognosis of patients. Increasing evidences clarify that autophagy-related genes (ARGs) make great differences to the generation and progression of LIHC, and may serve as prognostic biomarkers for LIHC. Methods We randomly divided the LIHC patients in The Cancer Genome Atlas (TCGA) into the training and testing group. Next, use the training group to perform univariate Cox, LASSO and multivariate Cox analysis to construct our prognostic index (PI) model for LIHC; use the testing group and the whole TCGA set to make internal validations; use the International Cancer Genome Consortium to make external validations; and use the whole TCGA, GSE14520 and Oncomine to exam the expression patterns of the five ARGs. Then, we performed the ROC curve as well as univariate and multivariate analysis to evaluate the independent prognostic prediction power of the PI model, and made nomograms to estimate 1,3,5-year survival rate of LIHC patients. Besides, we conducted functional enrichment analyses of differentially expressed ARGs with GO, KEGG and GSEA, and made drug sensitivity analysis for the PI model via the GDSC database. Results A novel PI model which was composed of five key ARGs ( ATG9A , EIF2S1 , GRID1 , SAR1A and SQSTM1 ) succeeded to be constructed. All the internal and external validations testified that the PI model could well distinguish high-risk patients from low-risk ones, with AUC values > 0.60. Further comparison analysis showed that the PI model was no less than some common prognostic factors. People can estimate the 1,3,5-year survival rate of individual LIHC patient with the nomograms. Additionally, we obtained 62 differentially expressed ARGs and studied the potential mechanisms or pathways. Furthermore, we also found some potential targeted drugs associated to the five ARGs for LIHC patients. Conclusions The novel five-ARGs PI model has great potential to serve as a diagnostic or prognostic biomarker and therapeutic target in LIHC, which may guide future clinical applications to some extent and improve the outcome of LIHC patients.