With the increasingly serious situation of water resources shortage, the shortage of agricultural irrigation water in some areas has resulted in reduction of crop and damages the farmers’ interests. Soybean is kind of crop with high water requirement. Once the water deficiency will directly affect the morphology and growth, the quality and the yield will be reduced. Because water status of soybean leaves can truly reflect the degree of soil water deficit, a tool for water content measurements is in great need. The strong attenuation of terahertz radiation in water makes it a contactless probe, which can be used to detect the water status of leaves quickly. As a result, terahertz spectroscopy technology was studied to rapidly and conveniently estimate water content in soybean canopy leaf, so as to monitor the health status in real time. Zhong-huang 13 soybean cultivars were cultivated in our experiment. In order to simulate the drought stress of different degrees in the field, 5 different gradients of flowering soybean were carried out: normal watering, mild drought stress and moderate drought, severe drought, more severe drought stress (accounted for 80%, 65%, 50%, 35%, 20% of the maximum water holding capacity in the field, respectively) and 3 repetitions were set per gradient. The artificial weighing method combined with the portable soil moisture measuring instrument was used to regulate soil moisture content to meet the requirements of the various water gradients . Then, the experimental soybean were transported to the laboratory, and the samples were scanned by terahertz time domain spectrometer. 18 canopy leaves for each gradient, a total of 90 samples were collected. It was divided into calibration set and prediction set at 2∶1 ratio. After obtaining the time domain spectral data of each sample, the absorption coefficient spectrum and the refractive index spectrum of each sample were calculated by the data processing method of Dorney and Duvillaret. The changes of time domain spectroscopy, absorption coefficient and refractive index with water drought stress were qualitatively analyzed. It was found that the peak value of time domain spectrum was decreasing with the degree of water stress decreasing, which was lower than the reference value. At the same time, there was a significant time delay. The number of absorption coefficient gradually decreased with the aggravation of drought stress, and the refractive index value the same decreased. Moreover, partial least squares (PLS) and multiple linear regression (MLR) were used to quantitatively study the correlation between time domain spectrum, absorption coefficient, refractive index spectrum data and leaf water content, respectively. The results showed that, terahertz was sensitive to differences of leaf water content. And the MLR model based on maximum and minimum values in time domain spectral performed the best, in which correlation coefficient (rP) and root mean square error of prediction set (RMSEP) were -0.939 3 and 0.049 5, respectively. This study showed that the application of terahertz technology in leaf water content estimation has good feasibility. It will provide a new detection tool and experimental basis for rapid monitoring of water content in soybean canopy and scientific water-saving irrigation management.