Abstract: Leaf chlorophyll content is one of the most crucial indicators to evaluate the crop growth status in agricultural production. Hyperspectral remote sensing can be expected for the non-destructive detection of the chlorophyll content in leaves. The original spectral reflectance can be mathematically transformed into the derived spectrum. The noise interference can be reduced to enhance the sensitivity of the spectral information to biochemical parameters. Different band features of the derived spectra can also exhibit varying sensitivity to biochemical parameters. The accuracy of the chlorophyll content estimation can be improved to fully combine the sensitive features of multiple derived spectra. In this study, multiple derived spectral features were combined to estimate the chlorophyll content in winter wheat. Field experiments were carried out on the canopy spectral reflectance and leaf chlorophyll content of the winter wheat at four key growth stages (joint, heading, flowering, and filling stage) over four consecutive years. The original spectral reflectance was transformed to generate four derivative spectral features, including the first derivative reflectance (FDR), standard normal variable (SNV) transformation, de-trend (DT) correction, and continuum removal (CR) transformation. A comparison was made on the sensitivity difference between the original spectral reflectance and the four derived spectral features to leaf chlorophyll content. The sensitivity spectral features of the chlorophyll content were selected to eliminate the recursive feature for both single and multiple spectral features. Four machine learning algorithms (random forest, stochastic gradient boosting, support vector, and kernel ridge regression) were combined to construct the chlorophyll content estimation models, according to the single and multiple derivative spectral features. The results indicated that the determination coefficient (R²) between derived spectral features and chlorophyll content exceeded that of the original spectral reflectance. In the range of 400~436 nm, the R² value of the derived spectral feature SNV was between 0.514 and 0.567, which was much higher than that of the rest. In the range of 437~752 nm, the R² value of the CR was between 0.330 and 0.707. And the value of R² was greater than 0.6 in the range of 478~725 nm. In the range of 725~850 nm, the SNV and DT shared the largest R² values (between 0.358 and 0.219) in the different band ranges. The maximum R² of the five spectral features for the OR, FDR, SNV, DT, and CR were 0.575, 0.612, 0.688, 0.423, and 0.707, corresponding to the wavelengths of 681, 629, 702, 712, and 697 nm, respectively. There was a high accuracy of the machine learning model that was driven by the derivative spectral features to estimate the chlorophyll content. The better performance was achieved, compared with the original spectral reflectance (thirty spectral features were used to combine with the support vector regression). The highest accuracy was achieved in the random gradient lifting regression driven by 30 first-order derivative reflectance. The optimal model (R²=0.841 and RMSE=3.309 μg/cm²) was also obtained with the support vector regression, compared with the original spectral reflectance (R²=0.814 and RMSE=3.602 μg/cm²). Furthermore, 25 derived spectral features were combined to reduce the number of input spectral features. The stochastic gradient lift regression model was further improved to increase the accuracy of the chlorophyll content estimation, including two CR and SNV, six DT, and fifteen FDR spectral features. The estimation accuracy was R²=0.847 and RMSE=3.246 μg/cm². Multiple derivative spectral features were combined to drive the machine learning algorithms. The finding can provide a strong reference to further improve the estimation accuracy of the chlorophyll content for a less complex model. The improved model can also be expected for the aerial and aerospace hyperspectral remote sensing observation platform for the large-scale, accurate estimation of chlorophyll content in winter wheat leaves.