Molecular design approaches typically rely on simplified thermodynamic property prediction models to be computationally tractable. In addition, the simplified prediction methods have to be trained on initial experiments which usually defines and limits the molecular design space. In contrast, quantum mechanics can accurately predict properties independent of experimental data, but a direct integration into molecular design approaches is computationally challenging. In this work, we therefore aim at integrating quantum mechanical information into computer-aided molecular design (CAMD) while still allowing for efficient computations. For this purpose, COSMO-CAMD is presented as framework for molecular design based on COSMO-RS. Optimization-based molecule design is achieved with the genetic algorithm LEA3D, which creates 3D molecular structure information as input for COSMO-RS. A hierarchical approach is developed employing two accuracy levels for quantum mechanics. Thereby, COSMO-CAMD allows for the computationally efficient design of novel molecules even for problems involving molecules with complex molecular interaction such as fructose. In two case studies the COSMO-CAMD framework is shown to successfully design novel promising solvents in liquid liquid extraction of phenol and hydroxymethylfurfural from water. (C) 2016 Elsevier Ltd. All rights reserved.