Recent work implicating the cannabinoid receptors in a wide range of human pathologies has intensified the need for reliable QSAR models for drug discovery and lead optimization. Predicting the ligand selectivity of the cannabinoid CB(1) and CB(2) receptors in the absence of generally accepted models for their structures requires a ligand-based approach, which makes such studies ideally suited for quantum-chemical treatments. We present a QSAR model for ligand-receptor interactions based on quantum-chemical descriptors (an eQSAR) obtained from PM3 semi-empirical calculations for a series of phenyl-substituted cannabinoids based on a ligand with known in vivo activity against glioma [Duntsch, C.; Divi, M. K.; Jones, T.; Zhou, Q.; Krishnamurthy, M.; Boehm, P.; Wood, G.; Sills, A.; Moore. B. M., II. J. Neuro-Oncol., 2006, 77, 143] and a set of structurally similar adamantyl-substituted cannabinoids. A good model for CB(2) inhibition (R(2)=0.78) has been developed requiring only four explanatory variables derived from semi-empirical results. The role of the ligand dipole moment is discussed and we propose that the CB(2) binding pocket likely possesses a significant electric field. Describing the affinities with respect to the CB(1) receptor was not possible with the current set of ligands and descriptors, although the attempt highlighted some important points regarding the development of QSAR models.