To determine the freshness change of Channel catfish fillets during freeze-thaw cycles and establishprediction model using near infrared spectroscopy, Near Infrared(NIR) reflectance spectroscopy was used to collect spectral data of Channel catfish meat during 5 times freeze-thaw processing, meanwhile the contents of ATP-related compounds of fish samples were obtained by HPLC (High Performance Liquid Chromatography) to calculate the freshness of K value, then the quantitative prediction model of K value was established based on the optimal wavelengths using PLS (Partial least squares regression) method and verified by t-test. The results showed that ATP content declined rapidly (P＜0.05) from 2.15 to 0.55 μmol/g after the first freeze-thaw cycle and then stabilized until the end of processing; IMP content increased from the initial value of 4.07 μmol/g to the peak value of 5.97 μmol/g (P＜0.05) after the first freeze-thaw cycle and then decreased gradually (P＜0.05); Hx content increased slightly until the 4th freeze-thaw cycle and then increased greatly (P＜0.01) from 4.29 to 19.65 μmol/g after the 5th freeze-thaw cycle. K value significantly increased (P＜0.05) throughout the freeze-thaw processing, and the catfish meat showed decompose after 3th freeze-thaw while the K value was 67.95%(overlimitation of 60%). The near-infrared spectral data was pre-processed by filter fitting method (SG) and Standard Normal Variate (SNV), which obtained the best the prediction result, with the model predictive value determination coefficient (R²) of 0.938 1, and root mean square errors of prediction (RMSEP) of 1.49. Therefore, the established model was capable to predict the freshness of fish samples.