With the rapid development of modern agriculture, the use of pesticides has been increasing in both kind and level, and the corresponding problems of pesticide residue have attracted wide attention. Trichlorfon is an organophosphate insecticide used to control a variety of pests and parasites in livestock in the agriculture. Excessive intake of trichlorfon in the human body can cause problems in immune function, and even life-threatening. Therefore, the detection of trichlorfon residues is very important. Due to the unique properties of perspectivity, safety and spectroscopic fingerprinting ability, THz waves have achieved significant advances and applications as a new detection technology, and have beenin the fields of defense, industry, semiconductor, communication, biological medicine, pharmaceutical, agroproducts, food etc. Compared with traditional pesticide detection methods, terahertz spectroscopy has the characteristics of easy operation, short time-consuming, low cost and non-destructive. In this paper, terahertz frequency domain spectroscopy (THz-FDS) is used to detect trichlorfon in soil. The spectrum of pure trilobite tablets was analyzed. It was found that trichlorfon had characteristic absorption peaks at 1.18, 1.55 and 1.91 THz. DFT model B3LYP and basis function 6-31G(d) were used to calculate the single isolated molecule of trichlorfon to explain the source of these absorption peaks and verify the accuracy of the experimental results. In addition, 24 spectra of different contents (0.5%, 1%, 5%, 10%, 20%, 30%, 40%, and 50%) of trichlorfon in the soil were collected. It was found that the absorbance increased with the increase of the content when the content is more than 5%, showing a good linear relationship. The 24 spectra were divided into a calibration set and a prediction with a ratio of 3∶1, and were modeled by partial least squares method. The model has a relatively high correlation coefficient (>0.993 04), lower corrected root means square error (<0.021 9), a predicted root mean square error (<0.024 6), and a cross-validation root mean square error (<0.028 6). This paper provides a new method for the qualitative and quantitative analysis of pesticide residues in soil, as well as a new idea for the detection of pollutants in soil.