Abstract: In order to quickly detect the nutritional components of the forage in the temperate desert steppe in Inner Mongolia, 85 mixed fresh grasses were collected in the experiment, and the partial least squares regression method was used to establish a near-infrared reflectance spectroscopy（NIRS） predicted model for dry matter（DM）, crude protein（CP）, acid detergent fiber（ADF）, neutral detergent fiber（NDF）, ether extract（EE） and crude ash（Ash）. The results showed that the determination coefficient（R~2） of calibration set of DM, NDF, ADF and Ash were 0.985, 0.728, 0.749 and 0.727, respectively. The R~2 of validation set were 0.848, 0.536, 0.673, 0.741, respectively. The ratio of performance to deviation for validation（RPD） values were 8.163, 1.899, 1.927, 1.878（all greater than 1.75）, and the prediction accuracy of the established calibration model was high, which could be used for accurate detection in actual production. The R~2 of calibration set of CP and EE were 0.195 and 0.536, respectively. RPD was 1.117 and 1.479（both less than 1.75）, respectively. The validation set R~2 were less than 0.49, so the accuracy of the model needs to be improved. The results indicated that NIRS could be used to evaluate the nutritional indexes of natural mixed forage.