Abstract: In order to further improve the estimation accuracy of cotton chlorophyll density by hyperspectral reflectance, canopy hyperspectral reflectance and chlorophyll density were recorded at four different growth stages of cotton in a field experiment. All two-band combinations (350 to 1100 nm) in the ratio type of vegetation index (RVI) and the normalized difference type of vegetation index (NDVI) were performed on raw spectral reflectance and the first derivative reflectance, and then the correlation between all two-band combinations and cholorophyll density were determined. The coefficients (r) were presented in matrix plots. Basing on the results of correlation analysis, the estimation models of chlorophyll density were established using linear regression and multiply stepwise regression methods, and then the predictive power of four predictors were analyzed, i.e. single narrow band raw reflectance and the first derivative reflectance, the established vegetation indices for chlorophyll density estimation, and the optimal band combination vegetation indices. Three main conclusions were obtained: 1) The performance of first derivative reflectance was evidently better than raw reflectance; 2) The precision and stability of estimation models based on vegetation indices were normally much higher than models based on single band or multiply bands; 3) Among four types independent variables, DR756 was the best candidate for single-band models, ratio index DR635/DR643 and normalized difference index (DR1055-DR684)/(DR1055+DR684) were the best among all band combination indices. In conclusion, the model based on DR635/DR643 obtained the most satisfied results for the estimation of chlorophyll density, and the correlation coefficient between estimated and measured chlorophyll density reached 0.821. The study will provide a reference for the better application of hyperspectral reflectance in chlorophyll density derivation.