ABO blood typing is the determination of four different blood groups: type A, B, AB, or O. Clinically approved ABO blood typing methods are suffering from expensive reagents and multiple time-consuming cross-referencing steps, creating the need for fast, sustainable, sensitive, and label-free technologies. Raman spectroscopy techniques have shown potential to distinguish biomolecules and blood components such as purified serum proteins, albumin, and globulin. In combination with machine learning tools, the accuracy and specificity of Raman spectroscopic measurements can be improved and adapted to clinical applications. This study presents a multivariate analysis of human-blood samples for ABO blood typing using Raman spectroscopy and support vector machine (SVM) classification. A custom-built NIR Raman spectroscopy setup with a 785 nm wavelength laser is coupled into an inverted microscope to collect Raman spectra from each blood sample. Donor samples are drawn from EDTA tubes into a fused silica microcapillary without dilution and sample preparation steps. Raman measurements from more than 270 donor samples are analyzed to get accurate blood typing predictions. The blood types are distinguished pairwise by an average AUC score of 0.94, showing great potential of the developed system for future blood typing applications.