Different origins have significant impact on the secondary metabolites of traditional Chinese medicine (TCM), so identification of origins is helpful for scientific and rational utilization of TCM. Additionally, the detection of active ingredient content is the main way to evaluate the quality of TCM. In this study, we established a rapidly method to identify the origins and detect active ingredient content of Panax notoginseng in order to provide some research bases for scientific, rational and specific utilization and rapid quality assessment of P. notoginseng. A total of 117 Fourier transform infrared (FTIR) spectra of P. notoginseng originated from five regions were collected. The discrete wavelet transform was used to process the original spectra in order to remove part of the high-frequency signal caused by noise while partial least squares discriminant analysis (PLS-DA) was used to screen the data with the contribution rate greater than one. Moreover, 70% of the 117 individuals were selected to form the training set by using Kennard-stone algorithm as well as the other 30% were used as prediction set. Training set data were used to establish the discriminant model of support vector machine and the cross-validation method was used for screening optimal parameters as well as the prediction data were utilized to verify the results of discriminant model. The pre-processing data to predict saponins content were processed by standard normal variable transform and discrete wavelet transform. Processed date of FTIR spectra were set as variable X and the total contents of four kinds of saponins in P. notoginseng samples measured by high performance liquid chromatography (HPLC) were set as variable Y. The orthogonal signal correction was used to remove the unrelated data of FTIR to saponins content of P. notoginseng samples. 80% of the individual data were selected as training set and the other 20% were utilized to form the prediction set. The partial least squares regression model was established by training set and the prediction set was utilized to verify the results of the model. The results showed that, (1) The optimal parameters c and g of support vector machine calculated by cross-validation was 2.828 43 and 0.062 5 respectively and the optimal accuracy of training set was 91.463 4%. (2) The support vector machine model was set as the optimal parameter and the accuracy of prediction set was 94.285 7% which showed a high accuracy. (3) The correlation coefficient (R2) and the root mean square error of estimation (RMSEE) of partial least squares regression model established by training set was 0.941 8 and 4.530 7, respectively. (4) The R2 and the root mean square error of prediction (RMSEP) of partial least squares regression model verified by prediction set was 0.962 3 and 3.855 9 respectively showing predictive value of saponins content was close to the value detected by HPLC. FTIR combined with support vector machine could effectively identify different origins of P. notoginseng. Orthogonal single collection and partial least squares regression could accurately predict the value of total four saponins content of P. notoginseng. It could provide a simple, rapid, non-destructive, high sensitive detection method for the quality control of P. notoginseng.