The spectrum of surface plasmon resonance shows different amplitudes and peak widths with theeffects of resonant intensity and position, so it's necessary to develop the self-adaptive methods to guarantee the accurate positioning of SPR formant. Based on the optimized Savitzky-Golay filter, this paper presents a method for SPR spectrum, which can optimize the parameters (polynomial degree and window size) according to the characteristics of the real time SPR spectrum, and minimizes the Stein's unbiased risk estimate of the mean squared error, and it can be quickly solved by Monte-Carlo methods. It is confirmed that SURE can accurately reflect the trends and the results of the true MSE. The experiment measures the SPR signals of sucrose under different concentrations and integration time through the SPR system. By using the proposed method of optimal parameters on the SPR signal filtering, the results show that the repeatability error of the position of SPR formant is smaller compared with the fixed parameters, and does not increase with the addition of noise levels.