A near-infrared (NIR) hyperspectral imaging technique is investigated for non-destructive determination of moisture composition of jujubes produced in Li ngwu.The hyperspectral images of jujubes over the spectral region between 900nm and 1700nm a re acquired for 128jujube samples and the difference between raw wavelength and savitzky-golay smoothing wavelength is obtained by partial least-squares regression (PLSR).The important wavelengths are selected using weighted β-co efficients of PLSR,and the R-PLSR,A-PLSR and KM-PLSR models are th en established using these feature wavelengths related with the spectral information by PLSR to predict moisture of jujubes.The results suggest that the raw wavelength is superior the to savitzky -golay smoothing wavelength.Compared with full wavelengths,the results show that the op timal wavelengths by R-PLSR,A-PLSR and KM-PLSR models have an excellent ability to predict the content of moisture in jujube and K M-PLSR model is superior to R-PLSR and A-PLSR models to predict moisture composition.Their coefficient of det ermination (R²) and root mean square error of prediction (RMSEP) are 0.793and 1.828,respectively.Hence,it′s possible to determine the moisture of jujube by hyperspectral imaging technique.The image of feature wavelengths abstracted f rom the hyperspectral image and the model using these optimal wavelengths can provide a good prediction of jujube′s moisture.