Utilization of hyperspectral remote sensing technology to estimate wetland plant leaf nitrogen content quantitatively in large area is important to monitor and diagnosis of physiological condition and growth trend of wetland vegetation. However, there are many limitations such as over-fitting of inversion model, indeterminate causal relationship between the selected bands and biochemical parameters,"multicollinearity" of selected bands in leaf nitrogen diagnosis remote sensing research. The total nitrogen content of leaves of typical wetland plants, Phragmites australis and Typha angustifolia, was selected as our study objects. These plants grow in South Wetland purification system in the Olympic Park in Beijing, a typical wetland using reclaimed water. The leaf reflectance spectra of main wetland plants were acquired by means of an ASD FieldSpec 3 spectrometer (350-2500nm). Leaf total nitrogen content was determined by Kjeldahl nitrogen measurement method after acquiring the leaf reflectance spectra. The method,"Give a cross correlation analysis", was used to build the correlations between leaf nitrogen content and the original spectrum, first derivative spectrum. Then, the selected bands were divided into some areas according to QiSeGuang spectral range and the interval of selected bands. Those bands which are high frequency、frequency more、ranked high frequency are chosen as representative bands of different areas and considered to be the optimal bands used to build the regression model. Finally,partial least squares was used to build inversion model. The accuracy of this model was tested with cross-validated coefficient of determination (R_cv²) and cross、|validated root mean square error (RMSE_cv).The results show that the first derivative transformation can effectively improve the sensitivity of the original spectrum leaf nitrogen content inversion, and fully reflect the sensitivity of near infrared wave band representing leaf total nitrogen content. The accuracy of regression model based on the first derivative and the partial least-squares was much higher than that of the original spectra. In the regression model of the reed, verification accuracy (R_cv²) reached 0.84, square root error (RMSE_cv) was 0.11, in the regression model of the cattail, verification accuracy (R_cv²) reached 0.66, square root error (RMSE_cv) was 0.13, which were the optimal models to estimate leaf total nitrogen content. The determination of parameters in "Give a cross correlation analysis" provided a scientific basis for building the model of eliminating "singularband" and reducing "multicollinearity" problem. Spectrum zone division provides a scientific basis for revealing the causal relationships between optimal band and biochemical parameters. And partial least squares regression method was used to avoid "multicollinearity" of selected bands. The result from this study can not only fill the gaps in the detection of leaf nitrogen using remote sensing, but also provide a strong scientific basis for the nitrogen content monitoring and management of urban wetlands using reclaimed water. At last, Partial least squares regression method was used to avoid "multicollinearity" of selected bands.