The cannabinoid receptors belong to the membrane-bound G protein-coupled receptor superfamily and are predominantly coupled via Gi/o proteins. The CB1 receptor subtype is known to play an essential role in analgesia, memory-impairment, spasmolysis and regulation of appetite. Selective agonists for CB2 receptors show antiinflammatory and analgesic properties in animal models [1]. Cannabinoid receptors are activated by three major groups of ligands, mammalian endocannabinoids, plant and synthetic cannabinoids (e.g. THC from the plant Cannabis sativa). Here we investigated the interactions between the cannabinoid receptors type 1 and 2 (CB1/CB2) with a novel set of coumarin derivatives. Rational drug design for many GPCRs is complicated by the lack of receptor crystal structures. To aid interpretation of the experiments we have therefore constructed a model for the CB1 and CB2 receptors based on homology to bovine rhodopsin (pdb-code 1U19) and performed ligand binding simulations for a family of 39 coumarin derivatives applying an all-atom docking protocol using Flexscreen [2]. We rationally designed, synthesized and tested a novel set of ligands to exploit a hydrophobic cavity in the vicinity of the docking pose of known ligands. All novel ligands showed improved, affinities with respect to their reference compounds, in good quantitative agreement between experiment and simulation, some of them reaching nanomolar affinities. Studying a series of site-specific mutations, we could computationally rationalize the receptor selectivity of specific compounds to either preferentially bind CB1 or CB2, respectively. [1] S.H. Burstein and R.B. Zurier (2009) Cannabinoids, Endocannabinoids, and Related Analogs in Inflammation, AAPS J. 11(1), 109-119. [2] H. Merlitz, B. Burghardt and W. Wenzel (2003) Application of the Stochastic Tunneling Method to High Throughput Database Screening, Chem. Phys. Lett. 370, 68-73.