By using confocal Raman spectroscopy, Raman spectra were measured in normal rat red blood cells, normal human red blood cells, STZ induced diabetetic rats red blood cells, Alloxan induced diabetetic rats red blood cells and human type 2 diabetes red blood cells. Then principal component analysis (PCA) with support vector machine (SVM) classifier was used for data analysis, and then the distance between classes was used to judge the degree of close to two kinds of rat model with type 2 diabetes. The results found significant differences in the Raman spectra of red blood cell in diabetic and normal red blood cells. To diabetic red blood cells, the peak in the amide ⅥCO deformation vibration band is obvious, and amide ⅤN—H deformation vibration band spectral lines appear deviation. Belong to phospholipid fatty acyl C—C skeleton, the 1 130 cm-1 spectral line is enhanced and the 1 088 cm-1 spectral line is abated, which show diabetes red cell membrane permeability increased. Raman spectra of PCA combined with SVM can well separate 5 types of red blood cells. Classifier test results show that the classification accuracy is up to 100%. Through the class distance between the two induced method and human type 2 diabetes, it is found that STZ induced model is more close to human type 2 diabetes. In conclusion, Raman spectroscopy can be used for diagnosis of diabetes and rats STZ induced diabetes method is closer to human type 2 diabetes.