Leaf spectrum of winter wheat infected with stripe rust was measured using integrating sphere method. Spectral characteristic of single leaf with different severity levels (SL) has a significant difference at yellow light region. Spectral reflectance increases with the increasing of the SL significantly at visible light region and middle-infrared 1350～1600 nm, of which the result is similar to the early research results under canopy condition, but is different at near infrared light region. The mechanism of spectral reflectance changing was demonstrated in this paper. The wavelength bands between 446 nm and 725 nm as well as 1380 nm and 1600 nm were selected out as sensitive bands region to the SL. The individual bands of 493 nm, 666 nm and 1430 nm with the highest correlation coefficient were modeled with SL. The Spectral Angle Index (SAI) was designed to do regression with the single leaf SL using individual band 666 nm with the highest correlation coefficient as well as 758 nm with the lowest correlation coefficient. The feature absorption peak of initial spectral reflectance was normalized quantitatively by the continuum removal method. It indicated that the Depth and the Area of the feature absorption peak were closely related to the single leaf SL. Meanwhile, the Absorption Area Index (AAI) model, which can invert the leaf SL accurately, has been designed in the paper. The results indicate that the reflectance spectral feature is significant and the leaf SL of winter wheat infected with stripe rust can be inversed by the leaf spectral data. The results also provide a good basis for further studying monitoring mechanism of winter wheat stripe rust by remote sensing data.