Tetracycline under the condition of NaOH could be degraded to iso tetracycline which has strong fluorescent characteristic, and the prediction model of tetracycline contents in duck meat was developed with the combination of synchronous fluorescence spectrum, wavelet de-noising, particle swarm optimization algorithm (PSO) and support vector regression (SVR), and it could realize the rapid prediction of tetracycline contents in duck meat and improve the accuracy of prediction model. In the process, 70 nm was selected as the optimum wavelength difference for the determination of tetracycline contents in duck meat by using parallel factor analysis (PARAFAC). Secondly, the db6 wavelet with 2 levels decomposition was used to reduce the noise of synchronous fluorescence spectrum, and the spectrum after wavelet de-noising was normalized, and 6 characteristic wavelengths were selected by using PSO. Lastly, the SVR model parameters (c, g) were optimized by using PSO. Furthermore, the performances of the models of PSO-SVR, PLS and PCR under the spectral condition of characteristic wavelengths selected by using PSO, and PSO-SVR under the spectral condition of full spectrum were compared. The experimental results showed that the predictive ability of the model of PSO-SVR under the spectral condition of characteristic wavelengths selected by using PSO was strongest, and the correlation coefficient and the root mean squared error of prediction were 0.952 0 and 17.6 mg·kg-1, respectively. This work proved that PSO could extract effectively the characteristic wavelengths of tetracycline in duck meat, and the model of PSO-SVR could satisfy the request of rapid determination of tetracycline contents in duck meat.