The solubility of drugs in solvents is fundamentally important for drug development and manufacturing. As the experimental measurements of the solubility are extremely laborious tasks, reliable prediction methods are highly required. We have employed the conductor-like screening model for real solvents (COSMO-RS) in predicting the solubility of drugs and drug-like compounds in various solvent systems. We also evaluated the salt effect on the solubility of caffeine using this method. The present results demonstrated that COSMO-RS has reasonably reproduced the experimental data and have proved that this method is generally available in predicting the solubility of drugs.