To establish a rapid determination method for the contents of protein and starch in buckwheat by applying Near-Infrared Reflectance Spectroscopy (NIRS), NIR spectra of 217 buckwheat samples were collected. Calibration models were established based on the spectra and chemical values by partial least-squares (PLS) algorithm with cross-validation. Analysis indicated that the models of protein and starch showed good prediction effect by using multiplicative scattering correction (MSC), with rank of 5 and 5, respectively, and frequency regions of 6803.9-6094.2 cm-1. The R2 of them were 0.9481 and 0.9167, RMSECV were 0.68 and 2.08, RPD were 4.39 and 3.46, respectively, with the coefficients in external validation greater than 0.96. This study has demonstrated the feasibility and reliability of using NIRS to rapidly determine the content of protein and starch in buckwheat.