Background: Genomic expression profiles have enabled the classification of head and neck squamous cell carcinoma (HNSCC) into molecular sub-types and provide prognostic information, which have implications for the personalized treatment of HNSCC beyond clinical and pathological features. Methods: Gene-expression profiling was identified in TCGA- HNSCC (n = 492) and validated with the Gene Expression Ominibus (GEO) dataset(n = 270) for which RNA sequencing data and clinical covariates were available. A single-sample gene set enrichment analysis (ssGSEA) algorithm were used to quantified the levels of various hallmarks of cancer. And LASSO Cox regression model was used to screen robust prognostic biomarkers to identify the best set of survival-associated gene signatures in HNSCC. Statistical analyses were performed using R version 3.4.4. Results: We identified unfolded protein response as the primary risk factor for survival（cox coefficient = 17.4 [8.4-26.3], P < 0.001）among various hallmarks of cancer in TCGA- HNSCC. And unfolded protein response ssGESA scores were significantly elevated in patients who died during follow up (P = 0.009). Kaplan-Meier analysis showed that patients with low ssGSEA scores of unfolded protein response exhibited better OS (HR = 0.69, P = 0.008). And we established an unfolded protein response-related gene signature based on lasso cox. We then apply the unfolded protein response -related gene signature to classify patients into the high risk group and the low risk group with the cutoff of 0.18. Adjusted for stage，age，gender, our signature was an independent risk factor for overall survival in TCGA cohorts (HR = 0.39 [0.28-0.53],P = < 0.001). In external independent cohorts, similar results were observed. In the validation cohort GEO65858, the patients with high unfolded protein response score showed longer survival (HR = 0.62 [0.38-1.0], P = 0.049). And adjusted for stage，age，HPV state, the multivariate cox regression analysis showed that unfolded protein response-related gene signature exhibited an independent risk prediction for overall survival in 270 patients with HNSCC (HR = 0.57 [0.35-0.94], P = 0.026). Conclusions: By analyzing the gene-expression data with bioinformation approach, we developed and validated a risk prediction model with unfolded protein response -related expression scores in HNSCC, which have the potential to identify patients who could have better overall survival.