Abstract: In this paper,visible-near-infrared hyperspectral imaging technology was used to predict the fat content of cold Tan sheep mutton in order to optimize the best prediction model. By measuring the fat content of 90 longissimus dorsi muscles of Tan sheep and collecting their spectral images,the spectral prediction models of partial least squares regression(PLSR)and principal component regression(PCR)in full band were constructed after different pretreatments of the original spectra.After operation to reduce the model number,the pretreatment effect on the full wave model of the optimal continuous successie projection algorithm(SPA),competitive adaptive reweighted sampling(CARS),variables combination population analysis(VCPA)and interval variable iterative space shrinkage approach(IVISSA),and wavelength space iterative shrinkage method,through these methods to extract the characteristic wavelength,fat content of spectral prediction model was constructed. The results showed that,the PLSR full-band model constructed by Normlize pretreatment had the best effect,and the related coefficient of calibration set(Rc)of the correction set model reached 0.921. PCR full-band model constructed by multivariate scattering correction(MSC)pretreatment had the best effect,and the related coefficient of calibration set(Rc)of the correction set model reached 0.850. In the process of extracting characteristic wavelengths,the interactive verification root mean square error(RMSECV)of IVISSA algorithm was the lowest,which was 0.0072. Compared with the PLSR model constructed by the other three algorithms,the Normlize-IVISSA-PLSR model had the best effect,and related coefficient of calibration set(Rc)and related coefficient of prediction set(Rp)were 0.931 and 0.754,respectively. The above research showed that it was feasible to predict the fat content of Tan sheep mutton by hyperspectral method. The results provide a theoretical basis for the development of on-line fast nondestructive testing system for cold Tan sheep mutton quality.