We aimed to develop a high-throughput deep DNA sequencing assay of cerebrospinal fluid (CSF) to identify clinically relevant oncogenic mutations that contribute to the development of glioblastoma (GBM) and serve as biomarkers to predict patients' responses to surgery. For this purpose, we recruited five patients diagnosed with highly suspicious GBM according to preoperative magnet resonance imaging. Subsequently, patients were histologically diagnosed with GBM. CSF was obtained through routine lumbar puncture, and plasma from peripheral blood was collected before surgery and 7 days after. Fresh tumor samples were collected using routine surgical procedures. Targeted deep sequencing was used to characterize the genomic landscape and identify mutational profile that differed between pre-surgical and post-surgical samples. Sequence analysis was designed to detect protein-coding exons, exon-intron boundaries, and the untranslated regions of 50 genes associated with cancers of the central nervous system. Circulating tumor DNAs (ctDNAs) were prepared from the CSF and plasma from peripheral blood. For comparison, DNA was isolated from fresh tumor tissues. Non-silent coding variants were detected in CSF and plasma ctDNAs, and the overall minor allele frequency (MAF) of the former corresponded to an earlier disease stage compared with that of plasma when the tumor burden was released (surgical removal). Gene mutation loads of GBMs significantly correlated with overall survival (OS, days) (Pearson correlation = -0.95, P = 0.01). We conclude that CSF ctDNAs better reflected the sequential mutational changes of driver genes compared with those of plasma ctDNAs. Deep sequencing of the CSF of patients with GBM may therefore serve as an alternative clinical assay to improve patients' outcomes.