The selection of solvents is one of the most important factors in the early stage of chemical reactor design due to significant effects of solvents on both reaction kinetics and thermodynamics. In view of a large number of potential candidates, systematic methods are strongly required for guiding the solvent selection procedure. Unlike conventional solvent selection methods, which employ empirical or experimental parameters, the present work proposes a novel solvent screening method using theoretical descriptors to quantify solvent effects. These solvent descriptors are determined from COSMO-RS quantum chemical computations and advanced statistical analysis techniques. A model is developed by correlating a limited set of experimental reaction properties (reaction rate or reaction selectivity) in various solvents with the corresponding solvent descriptors. This model yields satisfactory solvent rankings for essential target properties. Based on extensive model predictions with 136 common solvents, a few promising candidates showing the highest reaction rates or selectivities were identified and their electronic and structural properties were characterized. The reliability of the proposed method was validated with three different reactions. As the underlying COSMO theories are solely based on molecular structure information, the method is a powerful tool for fast solvent pre-screening without the need for extensive experiments. (C) 2013 Elsevier Ltd. All rights reserved.