We have investigated the potential of Raman microspectroscopy combined with supervised classification algorithms to diagnose a blood lymphoproliferative disease, namely chronic lymphocytic leukemia (CLL). This study was conducted directly on human blood smears (27 volunteers and 49 CLL patients) spread on standard glass slides according to a cytological protocol before the staining step. Visible excitation at 532 nm was chosen, instead of near infrared, in order to minimize the glass contribution in the Raman spectra. After Raman measurements, blood smears were stained using the May-Grünwald Giemsa procedure to correlate spectroscopic data classifications with cytological analysis. A first prediction model was built using support vector machines to discriminate between the two main leukocyte subpopulations (lymphocytes and polymorphonuclears) with sensitivity and specificity over 98.5%. The spectral differences between these two classes were associated to higher nucleic acid content in lymphocytes compared to polymorphonuclears. Then, we developed a classification model to discriminate between neoplastic and healthy lymphocyte spectra, with a mean sensitivity and specificity of 88% and 91% respectively. The main molecular differences between healthy and CLL cells were associated with DNA and protein changes. These spectroscopic markers could lead, in the future, to the development of a helpful medical tool for CLL diagnosis.