Raman spectra of normal white blood cell (WBC) and diabetes mellitus (DM) WBC in vivo are studied by laser tweezers combined with multivariate statistic analysis method. Raman spectrum and multivariate statistic model is founded by the principal components analysis (PCA). The result of Raman spectrum suggests that DM WBC and normal WBC could be differentiated with PCA. Diagnostic specificity and sensitivity are nearly 98%. The protein vibration bands at 1302 cm-1 in DM WBC mouse in vivo show that the concentration of protein in DM WBC is higher than normal WBC. The Raman intensities at O-P-O of DNA and amide bands of DM cell are significantly higher than normal cell, which shows that the changes in the comformation of O-P-O and the backbone and hydrogen bonds of protein, as well as the second comformation of protein.