Abstract: Meat and bone meal can be widely used as an important protein feed raw material in breeding industry, because of its high content of protein and biological value. However, the feed safety issues occurred frequently around the world, such as mad cow disease and sheep pruritus. There is an urgent need to develop an identification technology with a simple, reliable, reasonable, and efficient way to distinguish the specific species of meat and bone meal. In this study, Raman spectroscopy and chemometrics methods were used to establish a systematic approach for the identification of species in meat and bone meal. The proposed method was based on the difference in Raman spectral characteristics of bone protein. This is due mainly to the composition and structure of bone proteins vary with species, while the bone protein was also the dominated component of meat and bone meal. A total of 87 source-reliable samples of meat and bone meal were analyzed, including 23 porcine, 22 poultry, 20 bovine, and 22 ovine in the measurement. Given the extraction procedures of proteins were complex, the bone protein was enriched via the extracting bone particles, and then ground into powder. All the powdered bone particles were scanned by the Raman spectrometer from 400 to 3 600 cm-1, where the Raman characteristics of bone protein, including 3200-2800, 1800-1200 cm-1, and 900-800 cm-1, were selected for further analysis. A three-step protocol was established for discriminant analysis combined with partial least squares method. The first model was used for the identification of poultry and mammal (porcine, bovine and ovine) meat and bone meal, where six latent variables were selected to establish the partial least square-discriminant analysis (PLS-DA) model. The specific fingerprint characteristic bands were at 1659, 2930, 2852, 1246 cm-1, and 1455 cm-1, respectively. Both the sensitivity and specificity of discriminant models were achieved at 100%, indicating that the PLS-DA discriminant analysis model based on the Raman characteristic band of bone protein can be well suited to distinguish poultry and mammal (porcine, bovine and ovine) meat and bone meal. The second model was used for the identification of non-ruminant (porcine) and ruminant (bovine and ovine) meat and bone meal, where ten latent variables were selected to establish a PLS-DA model according to the Raman characteristic bands of bone protein. The fingerprint characteristic bands were located at 2852, 1659 cm-1, and 1246 cm-1, respectively. Both the sensitivity and specificity of discriminant model can also be 100%, indicating that the PLS-DA discriminant analysis model can be used to well distinguish non-ruminant (porcine) and ruminant (bovine and ovine) meat and bone meal. The third model was used for the identification of bovine and ovine meat and bone meal, where eleven latent variables were selected to establish a PLS-DA model. Their fingerprint characteristic bands were at 878, 853, 2930, 2852, 1246, 1455 cm-1, and 1659 cm-1, respectively. Both the sensitivity and specificity of discriminant model can be obtained over 93.8%, indicating that the PLS-DA discriminant analysis model can be used to well distinguish bovine and ovine meat and bone meal. The findings can be contributed much to the rapid development of a portable Raman spectrometer system for species-specific identification of meat and bone meal in feed industry.