Accurate prediction of pollutant dispersion is vital to the energy industry. This study investigated the Computational Fluid Dynamics (CFD) simulation of pollutant emission in a natural draft dry cooling tower (NDDCT) with flue gas injection. In order to predict the diffusion and distribution characteristics of the pollutant more accurately, Large Eddy Simulation (LES) was applied to predict the flow field and pollutant concentration field and compared with Reynolds Average Navier-Stokes (RANS) and Unsteady Reynolds Average Navier-Stokes (URANS). The relationship between pollutant concentration pulsation and velocity pulsation is emphatically analyzed. The results show that the flow field and concentration field simulated by RANS and URANS are very close, and the maximum value of LES is about 43 times that of RANS and URANS for the prediction of pollutant concentration in the inner shell of cooling tower. Pollutant concentration is closely related to local flow field velocity. RANS and URANS differ greatly from LES in flow field prediction, especially at the outlet and downwind of cooling tower. Compared with URANS, LES can simulate flow field pulsation with a smaller scale and higher frequency.