This paper established a rapid quantitative detection method for the content of potassium sorbate in orange flavoreddrink by using surface-enhanced Raman technique, which is based on the self-built laboratory Raman point scanning system. Through the comparative analysis of these Raman spectra and surface-enhanced Raman spectra of potassium sorbate under different states, such as standard potassium sorbate, aqueous solution of potassium sorbate and so on, identified 1 159.3, 1 398.9 and 1 653.0 cm-1 were characteristic Raman peaks of potassium sorbate, 1 648.4, 1 389.3 and 1 161.8 cm-1 were surface-enhanced Raman characteristic peaks of potassium sorbate. Confirmed this method had better repeatability by doing Raman shift peak intensity reproducibility experiment of the parallel samples and calculating the relative standard deviation (RSD) of the peak intensity. The results showed that the peak intensity average relative standard deviation RSD value was 9.88%. Gathered the surface enhanced Raman spectra of 33 orange flavored beverage samples from the concentration range of 0.180 7~1.706 g·kg-1 and all these Raman spectra did the pretreatments of S-G 5 point smoothing and Baseline removal of fluorescence background. Combined with the commonly used quantitative analysis method (Linear regression analysis method, multiple linear regression analysis method, and partial least squares regression analysis method), three prediction models which were based on these three different principles were established for the content prediction of potassium sorbate. By comparison, the multiple linear regression model which used three Raman shifts (1 161.8, 1 389.3 and 1 648.4 cm-1) had the smallest prediction error and highest model precision. In the multiple linear regression model, correlation coefficients of correction set and validation set (R2C and R2P) were 0.983 7 and 0.969 9, root mean square error of correction set and validation set (RMSEC and RMSEP) were 0.051 7 and 0.052 8 g·kg-1. And the relative standard deviation (RSD) of this model was 9.93% and the relative error (RPD) of this model was 5.06. By using the surface enhanced Raman technique combined with multiple linear regression analysis method, a more accurate and rapid prediction of potassium sorbate in orange beverage can be realized, which lays a technical foundation for the rapid monitoring of potassium sorbate in other foods.