The lipid oxidation of fresh chilled pork was evaluated rapidly and contactlessly by near-infrared hyperspectral imaging during different storage periods. The Gaussian filter smoothing (GFS), moving average smoothing (MAS), Savitzky-Golay convolution smoothing (SGCS), median filtering smoothing (MFS), multiplicative scatter correction (MSC), standard normal variable (SNV) correction and baseline correction (BC) were used to preprocess the reflectance spectra of the pork samples at 900~1 700 nm. After the seven kinds of pretreatments mentioned above, Partial least squares regression (PLSR) model was established to explore the quantitative relationship between spectral data and the 2-thiobarbituric acid (TBA) value. As a result, the GFS-PLSR model based on full 486 wavelengths of GFS spectra showed better performance in prediction (R_P=0.919, RMSEP=0.036 mg/100 g). Regression coefficients (RC) method, stepwise and successive projections algorithm (SPA) were used to select optimal wavelengths to simplify the GFS-PLSR model and improve the efficiency of prediction. It was indicated that the RC-GFS-PLSR model established with the 29 optimal wavelengths selected from GFS spectra by RC showed better performance in prediction (R_P=0.924, RMSEP=0.034 mg/100 g), similar to GFS-PLSR model. The overall results showed that the TBA value could be quantitatively predicted by using near-infrared hyperspectral imaging technology combined with RC method, and this could be applied to realize the rapid and contactless evaluation of lipid oxidation in pork.