Abstract: In order to improve the model of wheat canopy nitrogen content ratio determination using hyperspectral reflectance spectra, the optimal parameter combination of wavelet denoising was selected through orthogonal test (wavelet function: haar; decomposition level: 5; threshold option scheme: fixed form threshold；noise structure: unscaled white noise), and the partial least square (PLS) models were established with the denoising spectra of wheat canopy to compare the results of different pretreatment methods. It was found that the pretreatment method of wavelet denoising combined with first derivative could eliminate the background information of the original spectra most effectively, with root mean square error of calibration set (RMSEC) 0.260 and root mean square error of prediction set (RMSEP) 0.288, respectively. Then the pretreated spectra was analyzed using principal component analysis (PCA), and the top 6 principal components were used as the input variables for the least square support vector regression (LS-SVR) modeling. The RMSEC and RMSEP of LS-SVR model were 0.154 and 0.259, respectively, lower than that of PLS model, which indicated the LS-SVR model was more accurate. The results suggest that it is feasible to improve the accuracy of the model by eliminating the soil background information of original spectra with the pretreatment method of wavelet denoising combined with first derivative, and the LS-SVR algorithm is a preferred method of modeling.