Objective：To investigate the diagnostic value of DNA methylation levels of the host genes SOX1，JAM3，ZNF671，and DLX1 for cervical intraepithelial neoplasia grade 2 or worse（CIN2+）in women positive for human papillomavirus（HPV）16/18，and to construct a risk stratification model to optimize cervical cancer screening strategies. Methods：Cervical exfoliated cell samples were collected from HPV16/18-positive women attending the General Hospital of Eastern Theater Command between April 2023 and May 2024. DNA methylation levels of the four host genes were measured using bisulfite-converted pyrosequencing. Samples were grouped according to histopathological diagnosis（normal，low - grade histology，high - grade histology，cancer），and intergroup methylation differences were compared. Logistic regression was used to construct a predictive model for CIN2+，and its diagnostic performance was assessed in both training and testing datasets. Results：The methylation levels of all four host genes increased progressively with lesion severity and were significantly higher in the CIN2 + group than in the CIN1- group（P ＜ 0.001），demonstrating high diagnostic specificity. Among them，ZNF671 exhibited the best performance as a single biomarker，with an area under the curve（AUC）of 0.741. The combined model of ZNF671 and SOX1 achieved AUCs of 0.801 and 0.745 in the training and testing sets，respectively，with a specificity up to 87%，outperforming any individual gene. Conclusion：The DNA methylation levels of SOX1，JAM3，ZNF671，and DLX1 are closely associated with the severity of cervical lesions and exhibit potential value for triage. In particular，ZNF671 shows strong predictive performance for CIN2 + in HPV16/18 - positive women，and may serve as an auxiliary biomarker for identifying precancerous lesions. The combined model incorporating SOX1 further improves diagnostic accuracy.