Abstract: NIR hyperspectral imaging technology was investigated to rapidly evaluate Brochothrix thermosphacta in chilled chicken. After collecting hyperspectral images of fresh chicken and extracting reflective spectral information in the range of 900~1699 nm, the original spectra were pretreated by three methods including multivariate scattering correction (MSC), baseline correction (BC) and standard normal variable (SNV) correction. The quantitative relationship between spectral information and the reference value of Brochothrix thermosphacta in chicken samples was established by partial least squares (PLS) and multivariate linear regression (MLR). The optimal wavelengths were selected by PLS-β method, stepwise method and successive projection algorithm (SPA), respectively. The results showed that the full-band PLS regression model (F-PLS) built with BC spectra (485 wavelengths) performed better in predicting Brochothrix thermosphacta with R_P of 0.973 and RMSEP of 0.295 lg CFU/g.Twenty five optimal wavelengths were selected by PLS-β method to optimize the full wavelength PLS, resulting in the better performance with R_P of 0.931 and RMSEP of 0.434 lg CFU/g. In conclusion, NIR hyperspectral imaging technology had a great potential and could be used for rapid evaluation of the content of Brochothrix thermosphacta in chicken.