K value is an important freshness index. It indicates nucleotide degradation, which is usually employed to assess meat spoilage. High-performance liquid chromatography (HPLC) is a common analytical method to estimate K value. However, it is time-consuming, destructive, and is not suitable to monitor pork freshness rapidly, non-invasively and in real-time. The feasibility of Terahertz (THz) spectroscopy in predicting K value of pork non-destructively was studied. The THz spectra (0.2~2THz) of 80 pork samples with different freshness in the attenuated total reflectance (ATR) mode were acquired. Their K values were also measured by HPLC. Three models were established to predict K value, such as principal component regression (PCR), partial least squares regression (PLSR), and back propagation artificial neural network (BP-ANN) after the sample spectra were preprocessed by the first order derivative and filtered smoothly by Savitzky-Golay. Comparative research results showed that the nonlinear algorithm model of BP-ANN was the most superior way among three models whose root mean square error of prediction (RMSEP) was 14.36% and correlation coefficient (RP) was 0.75 in the prediction set. The THz spectral combined with BP-ANN model can be used to predict pork K value although it was not perfect. Compared with HPLC, the THz spectral was non-destructive, rapid and simple. The research would lay a theoretical foundation for developing portable THz inspection equipment based on the THz spectroscopy.