Cell-free DNA (cfDNA) molecules are short fragments generated through a non-random procedure. Hallmarks of cfDNA fragmentation patterns include a major peak at 166 bp and 10-bp periodicity below 143 bp, which characteristics had been hypothesized to correlate with the nucleosome structure. However, to date, the underline molecular mechanisms/regulators interacting with these nucleases are still unclear. In this study, we report a DNA methylation - nuclease preference - cutting end - size distribution axis, demonstrating the role of DNA methylation as a functional molecular regulator of cfDNA fragmentation. Hence, low-level DNA methylation would increase nucleosome accessibility and alters the cutting activities of nucleases during DNA fragmentation, which further leads to variations in cutting sites and size distribution of cfDNA. We further demonstrated the feasibility in enriching tumor-derived cfDNA in cancer patients through selection of molecules possessing ends discordant with nucleosomes, which inspired three cfDNA fragmentomic metrics that showed significant alterations in cancer patients; by incorporating machine learning, we further built high-performance diagnostic models and enabled accurate and sensitive cancer diagnosis.