Surface Plasmon Resonance (SPR) sensing technology has been widely used in biomedical diagnosis, chemical detection, food safety, environmental monitoring and other fields due to its high sensitivity, in situ and label-free and real-time detection capability. For a wavelength interrogated SPR sensor, its resonance wavelength is measured for obtaining the analyte concentration. Because the optical spectrum analyzer (OSA) used in the SPR sensor has different photoelectric responses at different wavelengths, the measured resonance wavelength will deviate from the actual value in some extent, resulting in the measurement error. A radiation-based relative reflectance correction method is proposed in this work, which consists of the following three steps: (1) establishing the instrument response function (IRF) of the OSA used; (2) using the IRF to convert the reflected intensity spectrum of the SPR sensor directly measured with the OSA into the reflected power spectrum; (3) using the reflected power spectrum to determine the reflectance spectrum of the SPR sensor. The obtained reflectance spectrum is not affected by the SPR sensor system. Compared with the conventional reflectance correction method that needs to measure the resonance and non-resonance spectra of the SPR sensor, our method makes the resonance peak in the reflectance spectrum have a narrower FWHM and a more symmetrical shape and thus allows for more accurately determining the resonance wavelength. Based on the goal of minimizing the fitting error, an order-adaptive polynomial fitting algorithm is performed for the resonance region to determine the resonance wavelength. Experimentally, the SPR spectra at different incident angles are measured, and the FWHMs of the reflectance spectrum obtained by our method are maintained at (100±10) nm, indicating a universal shape advantage of this correction method. Then 4 000 spectra are continuously collected, and their resonance wavelengths are calculated by the method, the results give a relative standard deviation of 0.007 8% and a processing speed of 12 ms per spectrum. This indicates that the method has good robustness against noise fluctuations and real-time processing of spectral data. Lastly, the SPR resonance spectra of NaCl solutions at different concentrations are measured, it shows a linear correlation coefficient of 0.998 5 between the resonance wavelength and refractive index of the solutions, and a Figure of Merit (FOM) two times higher than that of the conventional method. This indicates that the method has good reliability and improves the sensing performance from the level of the spectral processing algorithm. While the conventional method for improving FOM is from preparation technology, ours is easier to handle, and the effect is prominent. All the experimental results show that our SPR resonance wavelength determining the method that combines the radiation-based relative reflectance correction and the order-adaptive fitting algorithm has the characteristics of good reliability, fast calculation, high resolution, anti-noise and high FOM which can effectively improve the data processing and sensing performance of SPR sensor.