Six widely-used dense gas dispersion models (ALOHA, HGSYSTEM, SLAB, SCIPUFF, PHAST and TRACE) were used to calculated downwind chlorine gas concentrations following three railcar accidents. The accidents, where as much as 60 tons of chlorine were released, were located at Festus, MO (release from a ruptured 1-in. line while offloading), and Macdona, TX, and Graniteville, SC (release from a large hole due to an accident). Public sources such as National Transportation Safety Board (NTSB) reports were used to develop input data. Source emissions rates were refined based on source modeling with PHAST and TRACE and derivations using fundamental thermodynamic equations. No chlorine removal mechanisms, such as photolysis, chemical reactions, or deposition were accounted for. Given the same source emissions rates, the models' simulations of 10-min averaged cloud centerline concentration, at downwind distances ranging from 0.1 to 10 km, agree with each other within plus and minus a factor of two most of the time. For a very large release (Graniteville), the 2,000, 400, and 20 ppm contours are predicted to extend downwind about 1.3, 3.1, and 14 km, respectively from the source. There is also agreement among the models simulations of the plume widths and heights to the 2.000, 400. and 20 ppm contours. A major conclusion of the study is that estimation of the source or release term is important for reliable results, since the calculated chlorine concentrations are approximately proportional to the mass release rate. It must also be stressed that observed concentrations would be expected to be less than those calculated here, because removal by dry deposition at the surface and by chemical reactions in the plume have not been accounted for. (C) 2008 American Institute of Chemical Engineers Process.