Cerebral microbleeds (CMBs) are small brain vascular lesions detectable on magnetic resonance imaging and associated with elevated risks of stroke and cognitive decline. This study aimed to identify circulating protein biomarkers for early CMB detection and to elucidate molecular mechanisms across CMB subtypes. Serum from 43 patients with MRI-confirmed CMBs and 38 healthy controls was analysed using high performance liquid chromatography combined with liquid chromatography tandem mass spectrometry. We identified 151 differentially expressed proteins, and enrichment analyses highlighted inflammation, extracellular matrix remodelling, and lipid metabolism. Subtype-stratified enrichment further implicated MAPK/RAF/ERK signalling in deep lesions and insulin-like growth factor signalling in lobar lesions. Machine learning was used for candidate selection, and six proteins were prioritised, including matrix metallopeptidase 3 (MMP3), epidermal growth factor containing fibulin-like extracellular matrix protein 1 (EFEMP1), tissue inhibitor of metalloproteinases 1 (TIMP1), uromodulin (UMOD), nectin cell adhesion molecule 1 (NECTIN1), and ubiquitin A52 ribosomal fusion protein 1 (UBA52). Validation by enzyme-linked immunosorbent assay confirmed robust diagnostic performance for MMP3, EFEMP1, TIMP1, and UMOD with area under the curve values above 0.7, and EFEMP1 correlated positively with CMB burden. Western blotting further verified subtype-specific expression patterns and identified reticulocalbin 1 (RCN1), neogenin 1 (NEO1), and amyloid precursor-like protein 1 (APLP1) as discriminators between lobar and deep CMBs, with area under the curve values above 0.8. These findings define a serum proteomic landscape for CMBs, provide mechanistic insight into subtype-specific pathways, and support the development of non-invasive biomarkers for early diagnosis and subtype differentiation.