China is the worlds largest pork producer and consumer. The quality of pork affects the quality of protein intake for people. There needs to be more effective rapid non-destructive testing methods to cope with the huge testing needs. In order to rapidly determine volatile salt nitrogen (TVB-N), pH and moisture content of frozen pork and to propose a new method for pork quality testing, this paper uses near-infrared spectroscopy combined with chemometric methods to establish mathematical models for TVB-N, pH and moisture content of frozen pork. The NIR spectral data were collected and combined with chemometric tests to obtain the measured values of TVB-N, pH and moisture content. ~5 000 cm-1 region, while the absorption peaks around 8 600～8 450 cm-1 were significantly smaller than the other absorption peaks. The SPXY (sample set partitioning based on joint X-Y distances) algorithm was used to partition the data set into a training set and a test set in the ratio of 3∶1. The abnormal data were removed using Monte Carlo cross-validation (MCCV) and a partial least squares regression (PLSR) was used to establish The regression relationships of TVB-N, pH, water content and full-band spectral information were established by partial least squares regression (PLSR), and the raw spectra were pre-processed using data centering, Savitzky-Golay(S-G) first-order derivatization, S-G second-order derivatization, direct difference second-order derivatization and multiple scattering corrections to explore the appropriate pre-processing methods. The results show that the data centering, direct differential second-order derivation and second-order derivation achieve good experimental results, so the combination of competitive adaptive reweighted sampling (CARS), uninformative variables elimination (UVE), and multiplicative scatter correction (MSC) has been applied. The PLSR feature band model was developed and analysed by combining the competitive adaptive reweighted sampling (CARS), uninformative variables elimination (UVE) and successive projections algorithm (SPA). The results showed that the prediction models for TVB-N, pH and water content had excellent performance when the structures were data centered-CARS-PLSR, direct difference second-order derivative-CARS-PLSR and second-order derivative-CARS-PLSR, respectively, where the training set correlation coefficients RC were 0. 947 1, 0. 998 8 and 0. 997 1, respectively, and the root mean square errors (RMSE) were 1. 208 8, 0. 008 7 and 0. 001 5, respectively; the test set correlation coefficients RP were 0. 927 5, 0. 963 0 and 0. 945 9, respectively, and the RMSE were 1. 683 6, 0. 051 7 and 0. 005 6, respectively. In summary, it can be seen that the NIR band region is significantly correlated with pork TVB-N, pH and moisture content, and the use of NIR spectra can The NIR spectra can be used to predict TVB-N, pH and moisture content in pork accurately.