At least 120 non-olfactory G-protein-coupled receptors in the human genome are 'orphans' for which endogenous ligands are unknown, and many have no selective ligands, hindering the determination of their biological functions and clinical relevance. Among these is GPR68, a proton receptor that lacks small molecule modulators for probing its biology. Using yeast-based screens against GPR68, here we identify the benzodiazepine drug lorazepam as a non-selective GPR68 positive allosteric modulator. More than 3,000 GPR68 homology models were refined to recognize lorazepam in a putative allosteric site. Docking 3.1 million molecules predicted new GPR68 modulators, many of which were confirmed in functional assays. One potent GPR68 modulator, ogerin, suppressed recall in fear conditioning in wild-type but not in GPR68-knockout mice. The same approach led to the discovery of allosteric agonists and negative allosteric modulators for GPR65. Combining physical and structure-based screening may be broadly useful for ligand discovery for understudied and orphan GPCRs.