One approach to drug design involves determination of the structure of binding sites on target proteins to provide templates for ligand construction. Alternatively, random combinations of chemical groups can be used to generate diverse molecules for screening in the search for effective compounds. Here we report a strategy for developing potent ligands for proteins with multiple binding sites, which combines elements of both approaches: 'polymer-linked ligand dimers', in which two ligands are joined by a polymer chain of variable length. We find that polymer-linked ligand dimers containing two cyclic GMP moieties are up to a thousand times more potent than cyclic GMP in activating cyclic-nucleotide-gated channels and cGMP-dependent protein kinase. Each target protein responds optimally to a polymer-linked ligand dimer with a different average polymer length, even though their cyclic-nucleotide-binding sites are conserved. The tuning of polymer-linked ligand dimers indicates that each protein has a unique spacing of binding sites and provides an estimate of the distance between these sites. As optimal ligands are selected empirically, the polymer-linked ligand dimer strategy enables potent and selective agents to be identified without requiring previous structural information about the target proteins.