Thermodynamic methods based on COSMO (COnductor-like Screening MOdels), such as COSMO-RS (Real Solvent) and COSMO-SAC (Segment Activity Coefficient), represent significant and recent developments of solvation thermodynamics and computational quantum mechanics. These are a priori prediction methods based on molecular structures and a few parameters that are fixed for all of the compounds. They require no experimental data and rely on sigma profiles specific to each molecule as their only input. A sigma profile is the probability distribution of a molecular surface segment having a specific charge density. Generating sigma profiles by quantum mechanical calculations represents the most time-consuming and computationally expensive aspect of using COSMO-based methods. This article presents a free, web-based VT-2006 Solute Sigma Profile Database for large, pharmaceutical-related solutes, to supplement the published VT-2005 Sigma Profile Database for solvents and small molecules (www.design.che.vt.edu). Together, these databases contain sigma profiles for 1645 unique compounds, enabling the users to predict binary and multicomponent vapor-liquid equilibrium (VLE) and solid-liquid equilibrium (SLE), as well, as other thermodynamic properties. We validate the VT-2006 Solute Sigma Profile Database by solid solubility predictions in pure solvents for 2434 literature solubility values, which include 194 solutes, 160 solvents, and 1356 solute-solvent pairs. We also compare solubility predictions for mixed solvents to literature values for 39 systems. By comparison with experimental data, we find a root-mean-squared error (RMSE) of 0.7419 between experimental and predicted solute mole fractions (x(sol)) on a log(10) (x(sol)) scale for solubilities in pure solvents. This article also presents examples investigating the effects of conformational isomerism on solubility predictions of small, medium-sized, and large drug molecules. To provide better understanding of accuracy, we compare a priori COSMO-SAC solubility predictions, which use molecule-specific sigma profiles, to those by the non-random two-liquid segment activity coefficient (NRTL-SAC) model, which uses regressed molecule-specific parameters, for 17 solutes and 258 experimental solubility values. We find that NRTL-SAC, which contains regressed parameters based on experimental data, is a more accurate method for predicting SLE behavior than the COSMO-SAC model for many of the systems studied. Finally, this article presents a set of guidelines for applying the COSMO-SAC model for solubility predictions for new drug molecules when no experimental data are available.