Thiobarbituric acid reactive substances (TBARS) is the main chemical information that can effectively characterize the lipid oxidation degree in meat. A combination of hyperspectral technology with 2DCOS was investigated to develop the quantitative analysis model for assessing TBARS contents. The hyperspectral images of Tan mutton were collected in the spectral range of 400~1 000 nm. The regions of interest were manually set on the hyperspectral image of the, sample and the raw spectra data were extracted by ENVI 4.8 software. Partial least squares regression (PLSR) was established to correlate the spectra data with measured TBARS values. The kennard-Stone (KS) method was used to divide the whole data set to validate the calibration model. Savitzky-Golay (SG), de-trending and SG+de-trending were used to correct raw data for eliminating the interference information. The TBARS contents were regarded as a perturbation. Two-dimensional correlation spectra and their slice spectra were analyzed to determine the key variables associated with perturbation. Variable combination population analysis (VCPA), competitive adaptive reweighted sampling (CARS) and successive projections algorithm (SPA) was used to select variables from full spectra and 2DCOS analysis. PLSR algorithm was used to establish the hyperspectral quantitative analysis model of TBARS contents based on key variables. The result showed the PLSR model established by de-trending pretreatment presented good performance ($R_{\mathrm{C}}^2$=0.874, RMSEC=0.106 mg·kg^-1, $R_{\mathrm{P}}^2$=0.853, RMSEP=0.139 mg·kg^-1). The auto-correlation peaks related to TBARS values were observed at 579, 699, 756 and 867 nm. The wavelengths in the spectra range of 579~867 nm were the research area for detecting of TBARS contents. VCPA, CARS and SPA extracted 7, 16, 20, 8, 24 and 14 key variables from full spectra and 2DCOS analysis, respectively. Based on the accuracy and reliability of the obtained model, the model developed based on effective wavelengths selected by CARS from 2DCOS analysis could assess TBARS contents in an accurate and non-destructive manner. The quantitative model was: Y_(TBARS)=-0.15+2.99λ₅₈₈-7.01λ₅₉₃+7.45λ₅₉₈-6.14λ₆₀₃+7.06λ₆₁₂-8.25λ₆₂₂+2.64λ₆₃₁-4.18λ₆₃₆+13.91λ₆₄₆-11.3λ₆₅₅+12.64λ₆₇₅-8.51λ₆₈₄-7.81λ₆₈₉+1.08λ₇₀₃-2.54λ₇₁₃+5.47λ₇₂₇+6.62λ₇₄₂+5.69λ₇₅₁+2.48λ₇₇₅-1.93λ₇₈₀-6.95λ₇₉₀+7.09λ₇₉₉-3.56λ₈₀₉+1.82λ₈₁₉ ($R_{\mathrm{C}}^2$=0.857, RMSEC=0.113 mg·kg^-1). The research demonstrates that 2DCOS provide new insights into variable selection in spectra analysis. The combination of hyperspectral technology and 2DCOS is feasible for non-destructive monitoring of TBARS contents in mutton.