Abstract: Rapid and non-destructive identification of three varieties of beef(Holstein cow,Qinchuan cattle and Simmental cattle)was realized by visible/near-infrared hyperspectral imaging technology. Firstly,the original spectral date was pretreated and the sample set was divided.Then,the characteristic wavelengths were selected from the pretreatment spectral data by competitive adaptive reweighed sampling(CARS),successive projections algorithm(SPA)and uniformative variable elimination(UVE). K-nearest neighbor(KNN)and partial least squares discrimination analysis(PLS-DA)and support vector machine(SVM)discriminant models of beef were established,basing on full spectrum and characteristic wavelengths respectively. The results showed that the first derivative(FD)method was the optimal pretreatment method. The sample model divided by spectrum-physicochemical value symbiosis distance method(SPXY)had the best prediction performance. The number of the characteristic wavelengths selected by competitive adaptive reweighed sampling(CARS),successive projections algorithm(SPA)and uniformative variable elimination(UVE)were 24,17 and 19. The accuracy of the correction set and prediction set of the RBF-SVM model based on the characteristic wavelength,extracted by the CARS method were 100% and 98.82%,respectively. It was confirmed that using hyperspectral imaging technologies could obtain a better recognition effect of beef varieties. The study provided meaningful references for a rapid and non-destructive detection of the beef breeds.