In order to realize the on-line accurate measurement of turbidity of traditional Chinese medicine (TCM), the spectrum characteristics analysis method was used to establish the relationship between the spectrum characteristics and turbidity of the transmitted and scattered light signals. A near-infrared LED was employed as the light source of the sensor. The transmitted light and scattered light signals were converted into weak current signals by FDS100 photodiodes, then conditioned by a trans-impedance amplifier and a low-pass filter. The processed signals then were converted by A/D converters to digital signals. A MCU (STM32F405) was used to calculate fast Fourier transformation and turbidity. Finally, sixteen groups of standard turbidity solution (Formazin) ranging from 0NTU to 1000NTU were prepared using gradient dilution method to calibrate our sensor and make further verification, in addition, the sensor was applied for determination of the turbidity of angelica essential oil. The results show that the related coefficients of linear fitting of a ratio of the third and fifth harmonics components of the transmitted signal amplitude and the scattered signal amplitude are 0.9883 and 0.9946, respectively. And the minimum error and maximum error of our sensor are 0.471% and 3.768%, respectively. A good linear fitting degree of the turbidity of angelica essential oil is 0.99176, which satisfies real-time and on-line measurement requirements of essential oil extraction, concentration and drying. Therefore, the sensor has certain application filed for manufacturing and quality monitoring of TCM.