Recent researches have established that lncRNAs (long non-coding RNAs) could be exploited as new signatures for head and neck squamous cell carcinoma (HNSCC) diagnosis, prognosis, and treatment. Herein, HNSCC transcriptome data was abstracted from the Cancer Genome Atlas (TCGA) data resource, and uncovered immune linked lncRNAs through co-expression analysis. Besides, univariate along with Lasso penalty regression were employed to determine immune-linked lncRNA pairs with different expressions. We then compared area under the curve, calculated the Akaike information criterion (AIC) value of the receiver operating characteristic curve for 5 years, determined cutoff points, and established an optimal predictive model for identifying high- and low-risk HNSCC patients. Overall, we identified 40 differentially expressed immune-linked lncRNA pairs, 17 of which were incorporated in the Cox regression model. Using this model, we can more effectively stratify patients based on poor survival results, positive clinicopathological features, specific tumor immune invasion status, low chemotherapy responsivity, and high expression of immunosuppressive biomarkers. Our data illustrated that the immune-linked lncRNA pairs signature have clinical prediction value for HNSCC.