DNA methylation at age-related CpG (AR-CpG) sites holds significant promise for forensic age estimation. However, somatic models perform poorly in semen due to unique methylation dynamics during spermatogenesis, and current studies are constrained by the limited coverage of methylation microarrays. This study aimed to identify novel semen-specific AR-CpG sites using double-enzyme reduced representation bisulfite sequencing (dRRBS) and validate these markers, alongside previously reported sites and neighboring CpGs, using bisulfite amplicon sequencing (BSAS) to develop robust age estimation models. A methylome-wide association study was conducted on semen samples from 21 healthy Chinese men across three age groups, generating over 4 million CpG sites per sample at ≥ 5 × depth. Analysis of 721,840 shared CpG sites revealed that more than 95 % were not covered by conventional methylation microarrays. Differential methylation and correlation analyses identified 139 AR-CpG sites. A two-stage validation process using multiplex PCR-based BSAS was performed. In the first stage, 47 top dRRBS-identified AR-CpG sites, 26 literature-reported sites, and 242 neighboring CpGs were assessed in 129 semen samples (22-64 years), validating 31 dRRBS, 26 literature-reported, and 152 neighboring CpGs as age-related. The second stage examined 154 CpG sites in 247 samples (22-67 years), confirming 71 AR-CpG sites with |rho| >
0.50. Among these, chr2:129071885 (cg19998819) emerged as the strongest age-associated marker (rho = 0.81). Using the second BSAS dataset, age estimation models were developed with multiple linear regression and random forest (RF) algorithms within a repeated nested cross-validation (CV) framework (10-fold outer CV with 10-fold inner CV, repeated 10 times). The RF models demonstrated superior accuracy across feature subsets of 5-25 CpGs. The optimized 9-CpG RF model achieved an average root mean square error of 4.73 years (4.62-4.96, SD=0.10) and an average mean absolute error of 3.30 years (3.23-3.43, SD=0.06). This study demonstrates the utility of dRRBS for large-scale AR-CpG discovery and provides a robust age estimation model and a comprehensive reference database of semen-specific AR-CpG sites for forensic applications.