Near-infrared reflectance spectroscopy(NIRS) calibrations of crude protein and crude fiber in 152 alfalfa samples were developed by means of partial least squares(PLS) regression. Different interval wavelengths of 1, 2 and 5 nm and two kinds of samples(crude and fine) were set to collect spectral data and establish calibration respec tively. The accuracy and repeatability of calibrations were verified so as to compare superiority and inferiority. Results showed that the best validation result is the calibration by 2 nm interval wavelength and fine sample, and the correlation coefficients of calibration(R)were 0.97 and 0.94 for crude protein and crude fiber, and the SECV of these parameters were 0.42 and 0.78 respectively. The coefficients of determination for validation(R) were 0.96 and 0.92, and the SEP of these parameters were 0.43 and 0.79. The experiment showed that it is feasible to evalu ate the major quality traits of alfalfa by NIRS. All together, it provided a new model to verify the crude protein and fiber compositions of alfalfa.