A visible/near-infrared (400-1000nm) hyperspectral imaging technique is investigated for non-destructive determination of tendern ess composition of Tan-lamb mutton produced in Ningxia.The hyperspectral images of mutton over the spectral region between 400nm and 1000nm are acquired for 128mutton samples and the Savitzky-Gol ay (S-G) smoothing wavelength is selected.The 9important wavelengths are sele cted using weighted β-coefficients of partial least-squares (PLS) regression,and the models are then established usi ng these feature wavelengths related to the spectral information by PLS regress ion to predict moisture of jujube.At the same time,the multi-linear regression (MLR) model is established using 4feature parameters re lated to the image information.On the basis of these studies,a new partial least-squares regression (PLSR) model is establ ished by the fusion of spectrum and image feature information to predict mutton tenderness.The results show that the PLSR model of S-G smoothing wavelength is superior to the PLSR model of original spectrum,the PLSR model of the feature wavelengths is better than tha t of the all wavelengths,the PLSR model based on the fusion features of spectral and image information variables is superior to the PLSR model based on character istic variables of the single information,the correlation coefficient of prediction s et and the root mean square prediction error (RMSEP) are 0.89and 0.73,respectively.Therefore,the proposed method is feasible to predict quantitative detection of mutton tenderness,and provi des a theoretical basis for on-line detection of Ningxia′s Tan-lamb mutton quality in th e future.