CS₂ plays an important role in today’s chemical industry and other fields, while CS₂ fire pollution accidents are extremely harmful. It is necessary to study the fire pollution characteristics of CS₂ by studying the spectral radiation of CS₂ combustion flame. Has set up a platform CS₂ combustion flame spectrum test, this paper adopts from a complete set of VSR instrument calibration blackbody radiation sources, through infrared spectrum radiometer VSR test 5 cm, to 10 cm, 20 cm three scales CS₂ combustion flame spectrum, and through the thermocouple tested the whole combustion flame temperature under different combustion period, and installed over the flame to monitor the quality of flame in the combustion product concentration of flue gas analyzer. The flame temperature during the whole combustion period of CS₂, the flame spectrum and the composition information of combustion products at different combustion times and scales were measured. The test results show that the CS₂ flame mainly contains high temperature gases such as SO₂, CO₂, CO and H₂O, and the concentration of characteristic pollution product SO₂ is obtained. Due to the limited measurement resolution of existing spectrometers and the limited flame scale measured in laboratory experiments, to achieve online fire monitoring, it is necessary to establish a flame spectral radiation model to retrieve the pollutant concentration-related information in CS₂ fire. Based on the HITRAN database is near 2.7 μm for high temperature and water vapor emission peak, 4.2 μm for CO₂ emission peak, around 4.7 μm is CO faint emission peak, near 7.4 μm for SO₂ emission peak. We have gained CS₂, SO₂, CO₂, CO and H₂O gas in the absorption coefficient of the same temperature and calculate mixed gas transmittance and the emissivity, Combined with the gas radiation transfer equation and gas absorption coefficient equation, the flame spectral radiation model of CS₂ combustion is established. The spectral radiation model was used to invert the characteristic pollutant SO₂ under different combustion times and compared with the experimental data. The results show that this model has high accuracy and can to invert the concentration of combustion products quantitatively. The accuracy of inversion of SO₂ concentration at the time of 20, 40, 60 and 80 s is 89.5%, 82.5%, 85.6% and 86.5%, respectively. It lays a foundation for remote sensing monitoring and inversion of the concentration of combustion products in CS₂ large-scale fire.