Seven different storage period sausages were selected for nitrite content detection and corresponding spectral data collection，and uses Savitzky-Golary method to preprocess spectral data to reduce the noise of spectral data. Then based on the pre-processed spectral data, 29 characteristic wavelengths were extracted by partial least squares regression coefficient method. Finally, the detection accuracy of the prediction model of nitrite in sausages at characteristic wavelength and full wavelength were analyzed. The results showed that the prediction results of the regression model based on full wavelength were all higher than that based on characteristic wavelength, and the full-wavelength partial least squares regression model was superior to that of the principal component regression model, and the coefficient of determination of the accuracy of the partial least squares regression model was determined. The R² and root mean square errors were 0.982 9 and 0.059 2, respectively. The dissertation studies show that the spectral information at full wavelength is more suitable for the construction of hyperspectral detection model of nitrite content in sausage storage.