Abstract: To establish a rapid and accurate classification method for Astragalus herbs. Using a differential Raman spectrometer,88 batches of Astragalus herbs were examined under the conditions that the light source used a dualfrequency output (δλ ≤ 1 nm),a single-frequency laser with an output power of 250 mw,a linewidth of ≤ 0.06 nm,a wavelength of 785 nm,a spectral range of 180-2800 cm-1,and a scanning time of 3 seconds. According to the difference of Raman peaks,comparing the differential Raman spectra of the samples found that the samples could be classified into four kinds； the raw spectra were preprocessed by standard deviation standardisation method (Z-Score)； the discriminant analysis model was used for training,and the accuracy rates under prediction set and cross-validation were 95.5% and 81.8%,respectively； the accuracy rate of the test set in the established convolutional neural network model reached 100%； to further determine the importance of feature wavelengths for prediction,a random forest algorithm model was built using the same training and test sets,with accuracies of 90.8% and 85.0%,respectively. Differential Raman spectroscopy is able to detect different kinds of Astragalus herbs,and the general prediction model established is convenient and fast,which lays a foundation for the rapid examination and analysis of Astragalus herbs.