Abstract: Abstract: Chlorophyll is a good indicator of plant nutrition stress, photosynthetic ability and aging process. The use of hyperspectral remote sensing technology to monitor the chlorophyll status of soybean leaves is of great significance for soybean growth diagnosis and fertilization regulation. However, the information weakly correlated to chlorophyll is very easy to be neglected in the selecting process, which leads to the loss of effective information containing chlorophyll. Therefore, it is important to determine the weak information variables in leaf spectra so as to reveal the rules of hyperspectral response of chlorophyll. This study collected 80 sets of soybean leaf samples at seed-filling period in Jianghan Plain. The leaf hyperspectral data were measured by ASD HandHeld2 type spectrometer and SPAD (soil and plant analyzer development) value was measured by SPAD 502 chlorophyll meter in laboratory. The concentration gradient method was used to divide the whole sample set (80 samples) into a calibration set (54 samples) and a validation set (26 samples). Then the correlation between SPAD value and soybean leaf reflectance was analyzed, and as well as for every two spectral wavelength variables. The characteristic wavelengths of leaf spectrum were extracted from full bands based on iteratively retains informative variables (IRIV) method. Finally Partial Least Squares Regression (PLSR) model and Support Vector Machine (SVM) model were calibrated to estimate soybean leaf SPAD values by using characteristic variables and full spectral variables, respectively. The performance of the SPAD estimation model was tested using the determination coefficients (R2), root mean squared error (RMSE), and relative percent deviation (RPD). The results showed that the SPAD values of soybean leaves correlated strongly with the spectral wavelength variables in the visible bands, especially in the band of 500-650 nm and 690-730 nm, which were significantly negatively correlated with the spectral wavelength variables (P=0.01). The correlation between visible wavelength variables and near-infrared wavelength variables of soybean leaves was weak, and the correlation between internal variables of two bands was high, especially the collinear problem of near-infrared bands was prominent. Then 9 characteristic wavelength variables of soybean chlorophyll were determined based on IRIV algorithm. Among them, 5 strong information variables and 3 weak information variables were extracted in the visible bands, and 1 weak information variable was extracted in the near-infrared bands. The SVM model was better than the PLSR model. The R2 of validation set in PLSR model was 0.52, but in SVM model was higher than 0.59, so the SPAD value and the reflectance tended to be nonlinear correlated. IRIV algorithm can effectively determine the characteristic wavelength variables of SPAD value of soybean leaves. The estimation ability based on IRIV characteristic wavelength variables model was better than that based on full spectral variables model and strong information variables model. The IRIV-SVM performance was the best, and the R2 and RPD in validation set were 0.73 and 1.82, respectively. This study attempts to demonstrate the importance of weak information variables in the spectra and provides a theoretical basis for revealing the hyperspectral response mechanism of the leaves.