The Support Vector Regression (SVR) prediction model was established for the rapid detection of tetracycline contents in pork by the synchronous fluorescence spectroscopy combined with Competitive Adaptive Reweighted Sampling (CARS) method. The CARS was used to select tetracycline correlative variables of pork samples from spectral data and the optimum wavelength difference was set to be 65 nm. Then, the performance of three variable selection methods including CARS, Successive Project Algorithm (SPA) and Genetic Algorithm (GA) were compared. Finally, the SVR was used to establish the prediction model for tetracycline contents of the pork by using 16 selected variables. The results show that CARS after Multi Scattering Correction(MSC) processing is superior to SPA and GA and can select feature variables of full spectra efficiently. The prediction model of tetracycline by CARS-SVR is superior to the model established by original spectrum SVR, and the determination coefficient (R2) and the Root Mean Square Error of Prediction (RMSEP) in the CARS-SVR model prediction sets are 0.961 2 and 10.94 mg/kg, respectively. The results demonstrate that synchronous fluorescence spectroscopy combined with CARS-SVR is feasible to predict tetracycline contents of the pork, and CARS-SVR method can simplify the model efficiently and improve the prediction precision.