Optical tomography system based on modified simultaneous iterative reconstruction technique is designed to real-time monitor spatial distribution and diffusion process of smoke plume in a power plant. Concentration profiles, rather than just a point value, of smoke plume concentrations are the goal of this method, and the tomography algorithm is analyzed. According to incomplete projection data in the process of reconstruction and exiting noise interference, the modified simultaneous iterative reconstruction technique (SIRT) is adopted to extract the information of the trace gas concentration by a fan beam scanning and is compared to the conventional SIRT. Three evaluation indexes show that the reconstruction effect is the best by choosing proper relaxation factor, which reduces the index d to 0.044 from 0.435 and reduces the index r to 0.044 from 0.376 for 5500 iterations. Hence, the modified algorithm performs better in estimating the shape of the plume, this method is used to process the measured spectra in field campaign, the reconstruction results and the measured data are basically consistent, which is further confirmed by the experimental results.