A novel method is reported to discriminate human and animal blood by using Raman chemometric analysis. The phase information of Raman spectra was extracted with Hilbert transform and then analyzed with PCA and PLS to improve the accuracy of identification of human and animal blood compared with original spectra. The cluster analysis was made according to the principal component scores scatter plots of blood spectra data or its corresponding phase information. And the appropriate threshold value was set in the PLS-DA model in order to discriminate human and animal blood. The results show that the PCA model of the phase information can identify animal blood and human blood obviously and it exhibits higher recognition rate compared with PCA of original Raman spectra. The PLS-DA indicates that the optimal number of principal components for the phase information is 3, RMSEP and R2 are 0.044 3, 0.993 2, respectively. However, in the PLS model established with the original spectra, the optimal number of principal components is 6, RMSEP and R2 are 0.053 7, 0.990 1, respectively. This indicates that the PLS-DA model of the phase information can make less error by using less principal components. The RMSEP of PLS-DA model built by the phase information of Raman spectra is lower than that of the blood Raman spectra when taking the same number of fitting principal components. In conclusion, the complexity of the PCA and PLS models can be reduced and the recognition accuracy can be improved by extracting the phase information of Raman spectroscopy.