We study the feasibility of pork freshness grading using the spectra detected at different time with near infrared spectroscopy and establish the analysis model by using near infrared OPUS software. In order to reflect the pork freshness level better, we use SOM neural network clustering method to differentiate afresh the volatile base nitrogen national standard (Total Volatile Basic Nitrogen, TVBN) level, from 3 levels to 5 levels. In order to improve the prediction accuracy, based on the pretreatment by selecting a derivative+vector normalization method (13 smooth points), principal component analysis is used for dimension reduction before clustering analysis. It can reduce the prediction deviation, further improve the prediction accuracy of sample, reduce the prediction level deviation, and improve the prediction ability.