A "quasi-atomistic receptor model" refers to a three-dimensional receptor surface, populated with atomistic properties (hydrogen bonds, salt bridges, hydrophobic particles, and solvent) mapped onto it. In contrast to other 3D-QSAR approaches, an algorithm developed at our laboratory allows for the adaptation of the receptor-surface defining envelope to the topology of the individual ligand molecules. In addition, it includes H-bond flip-flop particles which can simultaneously act as H-bond donors and H-bond accepters toward different ligand molecules, binding to the surrogate within a pharmacophore hypothesis. Such particles mimic aminoacid residues able to engage in differently directed H-bonds at the true biological receptor. Ligand-receptor interaction energies are evaluated using a directional force field for hydrogen bonds and salt bridges. On the basis of a series of ligand molecules with individually adapted receptor envelopes, the software Quasar allows a family of receptor models to be generated using a genetic algorithm combined with cross-validation. Our concept has been used to derive semiquantitative structure-activity relationships for the beta 2-adrenergic, aryl hydrocarbon, cannabinoid, neurokinin-1, and sweet-taste receptor as well as for the enzyme carbonic anhydrase. The receptor surrogates for these systems are able to predict free energies of ligand binding for independent sets of test ligand molecules within 0.4-0.8 kcal/mol (RMS) of the experimental value.