Accurate prognosis prediction of head and neck squamous cell carcinoma (HNSCC) is vital for guiding clinical decision-making. DNA methylation (DNAm) plays important roles in the regulation of gene expression and abnormal DNAm is a hallmark of cancers. This study aimed to develop a prognostic model for HNSCC based on DNAm changes that can regulate gene expression. Based on The Cancer Genome Atlas HNSCC dataset, we constructed a novel four-CpG (cg02409878, cg01984743, cg23867673, and cg01995815) prognostic risk score (FCPRscore) model. The FCPRscore showed good generalizability on both TCGA and independent validation datasets, and achieved better performance compared with nine existing RNA/DNAm-based prognostic models. Furthermore, the FCPRscore is an independent prognostic factor and can enhance the predictive power of established clinical predictors. The low-risk group showed a more immune-active tumor microenvironment (TME) and more sensitive to immune checkpoint inhibitor therapy, while the high-risk group benefited more from chemotherapy. In addition, STK3, a gene regulated by cg02409878, significantly reduced proliferation and migration ability of SCC15 cells after belumosudil-mediated inhibition or shRNA-mediated knockdown. Our study proposed a robust prognostic model of HNSCC, FCPRscore, which could serve a prognostic factor and a promising tool to guide therapy decisions for HNSCC.