Abstract: 【Objective】 This project aims to study the correlation between the walnut leaves spectrum of 325-1,075 nm and the content of nitrogen in walnut leaves and explore the walnut leaf spectral data pretreatment and characteristic band screening methods, and establish a prediction model for the nitrogen content of walnut leaves.In order to realize rapid fertilization guidance in walnut production. 【Method】 First, a combined pretreatment method of multivariate scattering correction,Savitzky-Golay convolution smoothing filter and wavelet denoising were explored and established; then the feature bands were screened by the continuous projection algorithm; finally the predictive model of the nitrogen content in feature bands of walnut leaves were established by least squares regression. 【Results】 The results showed that the established combined pretreatment method had a better denoising effect on walnut leaf spectrum; Using the predictive model of nitrogen content in walnut leaves established in feature bands, the model's validation set determination coefficient R² reached 0.875, and the root mean square error RMSEP reached to 0.697,3 mg/g. 【Conclusion】 Compared with the full spectrum data, this established model reduces the influence of redundant data and noise, extracts the spectral information related to the effective components, and improves the quality of modeling.