Many lead compounds bind to serum albumin and exhibit markedly reduced efficacy in vivo as compared to their potency in vitro. To aid in the design of compounds with reduced albumin binding, we performed nuclear magnetic resonance (NMR) structural and binding studies on the complex between domain III of human serum albumin (HSA-III) and diflunisal, a cyclooxygenase inhibitor with antiinflammatory activity. The structural studies indicate that the aromatic rings of diflunisal are involved in extensive and specific interactions with hydrophobic residues that comprise the binding pocket in subdomain IIIA. The carboxylic acid of diflunisal forms electrostatic interactions with the protein similar to those observed in the X-ray structure of HSA complexed to myristic acid. In addition to the structural studies, NMR-derived binding constants were obtained for diflunisal and closely related analogues to develop a structure-affinity relationship for binding to subdomain IIIA. On the basis of the structural and binding data, compounds were synthesized that exhibit more than a 100-fold reduction in binding to domain III of HSA, and nearly a 10-fold reduction in affinity for full length albumin. Significantly, several of these compounds maintain activity against cyclooxygenase-2. These results suggest a rational strategy for designing out albumin binding in potential drug molecules by using structure-based design in conjunction with NMR-based screening.