Background: Cell-free DNA (cfDNA) from plasma has proved its power as a liquid biopsy analyte in many cancer types, except cancers from central nervous system (CNS). CfDNA from cerebrospinal fluid (CSF) better recapitulates mutations for primary CNS tumors or cancer brain metastasis. The characteristics of cfDNA from plasma, including concentration range, mutation frequency, length, end motif etc. have been thoroughly studied, while the features of cfDNA from CSF are yet to be investigated. Methods: We retrospectively reviewed records since 2021 from a clinical sequencing center, and selected samples from patients with non-small cell lung cancer. CfDNA from 30 CSF and 85 plasma samples was isolated, followed by NGS library preparation procedures, and enriched with a 680-gene panel. Fragmentation features including cfDNA length distribution and frequency of 6-bp end motif sequences were analyzed after removal of UMI. K-means clustering algorithm was utilized to analyze fragment motifs. Results: The mean concentration of cfDNA isolated from plasma and CSF was 13.46 ng/mL (1.28 – 96.40, 95%) and 26.82 ng/mL (2.80-382.00, 95%) respectively. A larger proportion of longer cfDNA fragments (>200bp) in CSF sample was observed, 37.8% in CSF VS 22.8% in plasma. Several end motifs were identified to be highly enriched in either sample type, and we found that sample types could be separated by K-means clustering from top 10 motifs. Conclusions: CSF is a less complicated solution system than plasma, regarding chemical constitutes. The cfDNA length distribution and motif patters are distinguishable in CSF and plasma. Our analysis demonstrated that the characteristics of cfDNA in CSF are obviously different from cfDNA in plasma, and the underlying mechanism remains to be clarified.