Objective: In this study

a high efficient and rapid method was used to identify the origin of herbal medicines in order to safeguard our country's economic interests in the international trade. Method: Ultraviolet(UV) spectroscopy combined with principal component analysis(PCA) and partial least square discriminant analysis(PLS-DA) was used to discriminate the Swertia davidi which collected from different origins and establish the prediction model to predict the accuracy of the regions. The spectra data were imported into UV Probe 2.34 software to compare the same part of S. davidi. Raw and pre-processed data(8 point smoothing

the first derivative and the second derivative) were imported into SIMCA-P 11.5 and the effect of discrimination of origins was compared by 3D score plot of PCA. Result: PCA indicated that the raw and 8 point smoothing data of leaves showed the best classification and the cumulative contribution rate of the first three factors was 98.8%. The other pre-processed methods could not obtain better identification and it may be related to the cumulative contribution value(the cumulative contribution rate of the data processed by the first derivative was 83.9%while the second derivative was 47.3%). Samples from Chongqing and Hubei could be distinguished with that of Hunan by the data of roots

but the samples of Chongqing and Hubei could not be separated. The model of PLS-DA may provide the basis of discrimination of more origins. The validation set was imported into the model developed by the training set and it proved that the model was feasible and effective. In PLS-DA

the correlation index of predictive value and true value in the training set was 0.985 and the RMSEE was 0.159. The correlation index of predictive value and true value after importing the validation set in the training set was 0.972 and RMSEP was 0.327. Both RMSEE and RMSEP were similar and less than 0.500. So the model had high reliability. Conclusion: UV spectra combined with PCA and PLS-DA can discriminate S. davidi from different origins and the predicted effect of the model was better. Furthermore

samples with unknown origins could also be distinguished.