Background: Breast cancer ranks as the second leading cause of cancer-related mortality among women globally. Early detection of breast cancer is crucial for improving patient outcomes and reducing mortality rates. Liquid biopsy, which relies on circulating tumor DNA (ctDNA) shed by breast tumors into the bloodstream, presents a promising non-invasive approach for early breast cancer detection. However, accurately distinguishing between benign breast abnormalities and malignant tumors remains a significant clinical challenge, as misdiagnosis can lead to unnecessary invasive procedures. Methods: Herein, we employed a multimodal analysis approach, namely SPOT-MAS (Screen for the Presence of Tumor by DNA Methylation and Size), to profile alterations in methylation and fragment length patterns of cell-free DNA (cfDNA) from 169 breast cancer-confirmed patients and 99 patients diagnosed with benign breast lumps including cysts, fibroadenomas, and fibrocystic changes. A robust machine learning model was constructed using these signatures to differentiate between breast cancer patients and individuals with benign lesions. Results: Our genome-wide analyses identified distinct profiles of methylation changes, copy number alterations, and end motifs in cfDNA, enabling discrimination between breast cancer patients and individuals with benign conditions. Notably, we observed a significant enrichment of Thymine and Adenine at cfDNA cleavage sites in breast cancer patients. Moreover, our target sequencing analyses uncovered distinct methylation patterns in the regulatory regions of multiple genes, including hypermethylation in GPR126 or hypomethylation in TOP1 or MAFB in breast cancer cfDNA. Our multi-featured model achieved an AUC of 0.92 (95% CI: 0.88–0.97), a specificity of 97.44% and sensitivities of 57.14% and 61.70% for stage I and stage II–III patients, respectively. Furthermore, our multimodal assay effectively differentiated multiple molecular subtypes of breast tumors from benign lesions, achieving the highest sensitivity of 71.43% with Luminal A, followed by Luminal B, Triple Negative Breast Cancer (TNBC), Luminal B-HER2, and HER2 with sensitivities of 66.67%, 60%, 58.33% and 55.56%, respectively. Conclusions: Our findings demonstrate the potential of cancer-specific methylation and fragmentomic patterns in plasma cfDNA as novel biomarkers for accurately discriminating between breast cancer and benign lesions. This capability reduces the false-positive rate and helps avoid unnecessary biopsies in current clinical practice.