The correlation of cotton leaf verticillium wilt severity level with raw hyperspectral reflectance, first derivative hyperspectral reflectance, and hyperspectral characteristic parameters was analyzed. Using linear and nonlinear regression methods, the hyperspectral remote sensing retrieval models of verticillium wilt severity level with remote sensing parameters as independent variables were constructed and validated. The result showed that spectral reflectance increased significantly in visible and short infrared wave band with the increase in the severity level, and this is especially significant in visible band. The raw spectral reflectance has the maximum coefficient of determination at 694 nm (R2 = 0.461 6) with severity level and the logarithm model constructed with reflectance at this point is the better one as compared to linear model. By the precision evaluation of retrieval models, the linear model with the first derivative reflectance at 717 nm as independent variable was proved to be the best, with R2 = 0.488 9, RMSE = 0.257 1, and relative error = 12.74%, for the estimation of verticllium wilt severity level of cotton leaf. The results provide a good basis for further studying monitoring mechanism of cotton verticillium wilt by remote sensing data, and have an important application in acquiring cotton disease information using hyperspectral remote sensing.