Abstract: To improve the accuracy and robustness of near infrared spectroscopy technique for rapid determination of jujube leaves’ nitrogen content, partial least squares (PLS) were used to establish the jujube leaf nitrogen content model in near infrared spectroscopy.Correlation coefficient of model was 0.799 and root mean square error was 0.055.The whole spectral region contains many spectral variables which has nothing to do with jujube leave nitrogen content.The existence of the redundant information reduces the prediction performance of the model.So the interval partial least squares (IPLS) combined with genetic algorithm and simulated annealing algorithm were used to extrat wavelength of the nitrogen content of jujube leaves.The nitrogen contents of jujube leaf samples were determined by Kjeldahl analysis method.The experiment selected 15 jujube trees; each tree selected 5 leaves as a test target.Spectral measurement instrument used in test is ASD spectrometer, and wavelength range of the machine is 350-2 500 nm, and the spectral resolution is 1nm.Whiteboard correction(standard whiteboard reflectivity is set to 1) was used before data collection, and each spectra sample were measured for 5 times, taking the average value as the relative reflectivity of the sample.The genetic algorithm combined with interval partial least squares method selected the four characteristic wavelengths 685, 689, 781, 783 nm.The nitrogen content of jujube leaves′ near infrared spectroscopy model was established according to these four wavelengths.Prediction correlation coefficient of model is 0.9175, and predicted root mean square error is 0.063.The near infrared spectroscopy model of jujube leaves′ nitrogen content was built based on seven wavelengths selected by simulated annealing algorithm.The correlation coefficient of model is 0.9301 and root mean square error is 0.052.Therefore, near infrared spectroscopy combined with the characteristics of the spectral selecting methods can effectively improve the accuracy of the model, making the model more practical.But the characteristics spectrum selecting methods are not universal.The model based on single wavelength variable selection is more sensitive, and it is more applicable to uniform samples.While the anti-interference of the model built based on wavelength interval selection is relatively stronger, and it is more suitable for heterogeneous samples.Therefore, the feature selection can be better used based on the combination of the state and the model.