The species identification of blood stains has important practical significance in criminal technology and inspection and quarantine. Raman spectroscopy provides an idea for the identification of bloodstain species. In this paper, human blood samples and blood samples of pig, chicken, duck, cow and mouse were collected and their Raman spectra were obtained. Savitzky-Golay method was used to smooth noise reduction, airPLS method was used for baseline correction, and 100~1 700 cm^-1 bands were selected for the experiment. The training set contained 600 sets of data, and the test set contained 300 sets of Raman spectral data. The first part of the experiment compared plS-DA, LDA, PCA+LDA, SVM and PCA+SVM. The accuracy of the test set was 84.0%, 49.3%, 78%, 83.0% and 85.7% respectively, which verified the effectiveness of the combination of the dimension-reduction algorithm and the SVM classifier. In the second part, three band selection algorithms of mutual information algorithm, genetic algorithm and equispaced combination were adopted. A comparative experiment was conducted in combination with the SVM classifier. The results showed that the combination of mutual information and the SVM algorithm had the best classification accuracy. When the number of band selection is 300, the accuracy of the three band selection algorithms combined with the SVM classifier is about 93%, which is much higher than the traditional classification method. The experimental results show that the spectral dimension reduction using a band selection algorithm can effectively improve the accuracy and robustness of the algorithm, and at the same time, make the identification of Raman spectral species more interpretable. The band selection algorithm determines the key band location of blood stain identification, which is also important for the design of a portable Raman system for law enforcement.