The quantum chemical COSMO-RS method was applied to analyze the equilibrium solubility of gaseous propane and propylene in a set of different room temperature ionic liquids in order to contribute to the design of olefin/paraffin separation technologies based on reactive absorption onto ionic liquids with dissolved silver salts media. First of all, the predictive capability of COSMO-RS was evaluated through a comparison of estimated values with a wide range of solubility experimental data; next a further optimization step based on the comparative analysis of predicted and experimental values of the Henry's law constant of each solute in different ILs was developed to improve the accuracy of the calculations. Afterwards, the optimized COSMO-RS approach was applied to select the most suitable RTILs for C3H6/C3H8 separation based on driving a computational screening of 696 RTILs. Results highlighted that small and symmetric fluorinated inorganic anions such as PF6- or BF4- provide lower solubilities for both hydrocarbons, but on a whole this results in higher separation selectivities. With regard to the structure of the cation, ILs based on imidazolium, pyridinium and pyrrolidinium cations provide similar properties. However ILs based on monosubstituted butyl ammonium have much lower solubilities for both gases and at the same time higher equilibrium selectivities for propylene. Also it was gathered that less and shorter alkyl chains in the cation also improve the selective separation of these mixtures. Finally, as the separation process is enhanced by the presence of Ag+ cations into the ionic liquid, the solubility of 8 commercially available silver salts was qualitatively related to the excess enthalpy of Ag+-IL in solution predicted by COSMO-RS in order to select a silver salt suitable to be dissolved in the selected RTIL. Thus, this work reports for the first time the use of a predictive tool in order to facilitate the design of innovative separation processes by reactive absorption in a Ag+-IL media. (C) 2013 Elsevier B.V. All rights reserved.