A new model is developed to predict Hg-0, Hg+, Hg2+, and Hg-p in the post-combustion zone upstream of a particulate control device (PCD) in pulverized coal-fired power plants. The model incorporates reactions of mercury with chlorinating agents (HCl) and other gaseous species and simultaneous adsorption of oxidized mercury (HgCl2) on fly ash particles in the cooling of flue gases. The homogeneous kinetic model from the literature has been revised to understand the effect of the NO + OH + M <-> HONO + M reaction on mercury oxidation. Because it is a pressure-dependent reaction, the choice of proper reaction rates was very critical. It was found that mercury oxidation reduces from 100 to 0% while going from high- to low-pressure limit rates with 100 ppmv NO. On the basis of the revised chemistry of SO2, NOx, HCl, Cl, and C/H/O, the mercury reaction mechanisms of Niksa et al., Qiu et al., and Widmer et al. were compared to the Sliger et al. furnace data on mercury speciation. The rate constants proposed by Niksa et al. were found to be in reasonable agreement with experimental data and, therefore, chosen for further model development. The heterogeneous model describes selective in-duct Langmuir-Hinshelwood adsorption of mercury. chloride on ash particles. The heterogeneous model has been built using Fortran and linked to Chemkin 4.0. This way, the simultaneous formation of HgCl2 in the gas phase and capture on ash is ensured. The final predictions of elemental, oxidized, and particulate mercury were compared to mercury speciation from power plant data. Information collection request (ICR) data were used for this comparison. The model results follow very similar trends compared to those of the plant data; however, quantitative deviation was considerable. These deviations are due to the errors in the measurement of mercury upstream of PCD, lack of adsorption kinetic data, accurate homogeneous reaction mechanisms, and certain modeling assumptions. The model definitely follows a new approach for the prediction of mercury speciation, and further refinement will improve the model significantly.