Abstract: In order to detect the contents of nitrite in commercial mustard by Fourier-Near infrared spectroscopy (NIRS), the nitrite contents of the mustard sample was determined according to GB5009.33-2016, and then sodium nitrite was added to the mustard sample to prepare a 60-sample calibration set with a sodium nitrite concentration ranging from 0.122 to 39.0875 with a concentration gradient of 0.66.A Fourier NIR spectrum curve corresponding to the 10 sample prediction set acquisitions was used to match the spectral information with actual measurement values, and the calculation analysis was performed by TQ analyst modeling software.The calculation and analysis results showed that the optimal modeling method was the combined preprocessing of the first order differential (1D) and Savitzky-Golay filter smoothing. A comparative analysis of spectral models established by partial least squares (PLS) and principal component regression (PCR) nitrite samples.The data showed that the partial least squares (PLS) nitrite component model had better stability and prediction ability, internal cross-test positive mean square error (RSMECV), cross-validation coefficient (R_c), external prediction root mean square error (RMSEP), prediction coefficient (R_P) correlation coefficient (r) were 0.0310, 0.9925, 0.0141, 0.9720 and 0.9378.After the F test and the t-test, there was no significant difference between the results measured by the national standard and the national standard.The NIRS test was rapid, non-destructive and convenient, and could be used for the quantitative detection of nitrite residues in commercial mustard.