This paper presents a new approach for predicting solvent effects on esterification reactions of industrial importance in the field of biocatalysis. The COSMO-RS method has been used to calculate the activity coefficients of the chemical species involved in various reactions, carried out in different solvents. For comparison we also used the traditional UNIFAC method. Three lipase-catalyzed esterifications were considered: (1) 1-dodecanoic acid with menthol in n-hexane, n-heptane, cyclohexane, 2,2,4-trimethylpentane, toluene, acetonitrile, and 2-methyl-2-butanol; (2) 1-dodecanoic acid and 1-dodecanol in n-hexane, n-heptane, cyclohexane, 2,2,4-trimethylpentane, and toluene; and (3) glycerol and n-octanoic acid in acetonitrile, benzene, and toluene and in the neat reaction mixture (without any solvent). Predicted activities were used to calculate the thermodynamic equilibrium ratio. This should be independent of medium, and the variation in COSMO-RS values is at most 9-fold (corresponding to a DeltaG degrees of about 5.5 kJ/mol, which would still be a very useful prediction) and often only 2-fold (corresponding to less than 2 kJ/mol or 0.5 kcal/mol, therefore comparable with experimental error). UNIFAC is weaker, especially when important roles are played by conformational freedom, intramolecular interactions, strong polar effects, and charge distribution of molecules in the mixture. The relative percent deviations from the mean of equilibrium constants in different solvents range between 17 and 49 for COSMO-RS versus 32 to 65 for UNIFAC. The COSMO-RS method opens up new perspectives for the development of theoretical models for solvent selection with general applicability.