Abstract: Near-infrared hyperspectral imaging(900~1700 nm)combined with chemometrics algorithms was investigated to rapidly predict the concentration of soy protein isolate in minced beef.The minced beef samples with different concentration of soy protein isolate(2%~30%,at interval of 1%,w/w)were firstly prepared. The hyperspectral images of the samples were collected and the spectral data within the images were then extracted. Based on the spectral information,the predicted models were established by partial least square regression(PLSR)and multiple linear regression(MLR). To improve model efficiency and reduce the collinearity of spectral data,optimal wavelengths were selected by regression coefficients(RC)and successive projection algorithm(SPA). The results demonstrated that the RC-PLSR model and RC-MLR model built with 22 optimal wavelengths selected by RC method performed better than the SPA-PLSR model and SPA-MLR model built with 21 optimal wavelengths selected by SPA method. The RC-PLSR model had best performance with r_P of 0.95,RMSEP of 2.73% and RPD of 3.32.The whole study revealed that near-infrared hyperspectral imaging combined with chemometrics could be used for rapid and non-destructive detection of the soy protein isolate concentration in minced beef.