In order to inspect accurately the condition of early wheat powdery mildew, and also to provide technical support for spraying pesticides, in this study hyperspectral imagery data of different disease severity for wheat leaves were collected at the early infection stage. Firstly, the leaf area and lesion area were segmented by image features, and then the disease severity was calculated quantitatively. Secondly, the Relief-F algorithm was introduced to select the most sensitive band and band difference, on the basis, the powdery mildew disease index(PMDI) was calculated. Moreover, the correlations between disease index(DI) and 11 vegetation indices(Including PMDI index) were analyzed, it was found that the PMDI index has the highest coefficient of determination(R2=0.839 9) and the lowest root-mean-square error(RMSE) which is 4.522 0. It was better than that of other disease vegetation indices, in which the result of normalized difference vegetation index(NDVI) was the highest, the determination coefficient is 0.777 1 and the RMSE equals 5.336 4. Finally, the support vector regression(SVR) models of PMDI and NDVI indexes were established, respectively, to further compare the retrieval performance for disease severity of early wheat powdery mildew. The result shows that the prediction model of PMDI index is better than NDVI index, the R2 is 0.886 3 with RMSE=3.553 2. It can be concluded that the developed method can effectively realize nondestructive diagnosis of early wheat powdery mildew, and provide important help for the spraying and disease control.