Background: Colorectal cancer (CRC) remains a leading cause of cancer related mortality worldwide. We utilized cell-free DNA (cfDNA) methylation and fragmentation characteristics of selected cancer-related biomarker regions and applied tumor-derived signal deduction and a machine learning algorithm to refine a blood test for the early detection of CRC. Methods: This was a prospective, international (Spain, Ukraine, Germany and USA [part of NCT04792684 study] population), observational cohort study. Plasma samples were collected either prior to a scheduled screening colonoscopy or prior to colonic surgery for primary CRC. 95 cfDNA samples from 48 early stage (I-II), 47 late-stage (III-IV) CRC patients (mean age 65 [48-83], female 45%, distal cancers 51%) and 204 age, gender and country of origin matched colonoscopy-checked controls were analyzed. 79 of the control patients had a negative colonoscopy finding (cNEG), 96 had benign findings of diverticulosis, hemorrhoids and/or hyperplastic polyps (BEN), 29 had non-advanced adenomas (NAA). Samples were analyzed utilizing previously described hybrid-capture based sequencing methodology. Panel of targeted biomarkers was previously identified through tissue- and plasma-based discovery and further narrowed down through cancer-related biological pathways analysis workflow. Individual cfDNA fragments belonging to each biomarker region were scored for cancer-specific signal. Finally, calculated scores were used in prediction model building and testing for establishing panel accuracy for CRC detection. Results: Prediction model utilizing a panel of methylation and fragmentation scores originating from cfDNA biomarkers belonging to relevant cancer development and progression related pathways, such as axonal guidance, ephrin receptor signaling, epithelial-mesenchymal transition and FGF signaling, correctly classified 92% (87/95) of CRC patients. Sensitivity per cancer stage ranged from 91% (21/23) for stage I, 92% (23/25) for stage II, 91% (30/33) for stage III and 93% (13/14) for stage IV. Fragmentation signals contributed most to early-stage cancers (I-II), while methylation signals were more significant for late stage (III-IV) detection. Specificity of the model was 94% (199/204), with 97% (28/29) NAA, 91% (87/96) BEN and 96% (76/79) cNEG patients correctly identified. Lesion location, gender, age and country of origin were not significantly correlated to prediction outcome. Conclusions: Use of methylation and fragmentation characteristics of cancer-related cfDNA regions, combined with a machine-learning algorithm is highly accurate for early-stage (I-II) CRCs (92% sensitivity at 94% specificity). The study is being further expanded on larger cohort for validation of a highly accurate and minimally invasive blood-based CRC screening test.