Abstract: In order to ensure feed safety, it is necessary to detect the meat and bone meal contents in concentrate supplement. The method of near infrared reflectance spectroscopy (NIRS) for predicting meat and bone contents in concentrate supplement was explored. The NIRS calibration models were developed from a calibration set of 123 samples by partial least squares (PLS) regression. The initial spectrum was pretreated by scatter correction, Savisky-Golay smoothing, first derivative and second derivative, respectively, and the methods of multiple scatter correction, Savisky-Golay smoothing and the second derivative were chosen as the best pretreatment based on the biggest coefficient of determination in calibration(R²) and the smallest standard errors(RMSEC). The R² and RMSEC were 0.9751 and 0.437, respectively. Thirty-four samples were chosen as validation set, the coefficient of determination in validation(r²) and standard errors(RMSEP) were 0.9749 and 0.420, and the absolute errors and mean relative errors were 0.3259 and 13.89%, respectively. It is concluded that NIRS can be used as a method to detect meat and bone meal contents in concentrate supplement.