Three Raman characteristic peaks of sodium benzoate located in 843.5 cm-1, 1007 cm-1 and 1605 cm-1 are identified by using self-built Raman spectroscopy detection system. Using silver colloidal prepared by reduction of silver nitrate with sodium citrate as surface enhancer, we develop a rapid detection method for sodium benzoate in commercial carbonated beverage. The surface enhanced Raman spectra of 44 carbonated beverage samples with different sodium benzoate concentrations are randomly divided into 32 calibration sets and 12 validation sets, and all Raman spectra are carried out the pretreatments of Savitzky-Golay smoothing and baseline removal of fluorescence background. Simple linear regression analysis method, multiple linear regression analysis method, partial least squares regression analysis method, principal component regression analysis method and support vector machine regression analysis method are used to build models. The results show that the optimal model is binary linear regression analysis model based on the sodium benzoate characteristic peaks at Raman shift of 843.5 cm-1 and 1605 cm-1. In the binary linear regression model, correlation coefficient of validation set is 0.9603, root mean square error of validation set is 0.0867×10-3. The rapid and quantitative detection of sodium benzoate in commercial carbonated beverages can be realized by the binary linear regression analysis model, and this provides technical support for the real-time online detection of the content of sodium benzoate in food.