The chemical properties of sodium nitrate are stable, and the Raman characteristic peaks of sodium nitrate and potassium sorbate can be separated completely. The Raman spectrum of potassium sorbate, a preservative commonly used in food, is calibrated with Raman spectrum of sodium nitrate, whose characteristic peak at Raman shift 1053 cm -1 of sodium nitrate with mass fraction of 0.1 is used as internal standard peak. We calculate the relative ratios of the intensity of internal standard speak and characteristic peak intensities of the same concentration sodium nitrate in 49 samples respectively, use the relative ratios to correct characteristic peak intensities of potassium sorbate in 49 samples, and establish the quantitative prediction model of potassium sorbate by linear regression. The results show that the correlation coefficients of calibration set and prediction set significantly improve after correction. The squares of correlation coefficients of potassium sorbate at 1399 cm -1 characteristic peak are 0.9885 and 0.9865, the root mean square errors are 3.0384×10 -3 and 3.7643×10 -3 for calibration set and prediction set, respectively. Based on the optimal prediction model, we predict 18 new samples. The square of correlation coefficient of the predicted and real values is 0.9799 and root mean square error is 4.8702×10 -3, which indicate that the internal standard method of sodium nitrate can effectively reduce the influences of detection instrument, detection environment, and human factors on the Raman peak intensity of potassium sorbate, and can improve the accuracy of the predicted model.