Derivatives of the nonselective excitatory amino acid antagonist kynurenic acid (4-oxo-1,4-dihydroquinoline-2-carboxylic acid, 1) have been synthesized and evaluated for in vitro antagonist activity at the excitatory amino acid receptors sensitive to N-methyl-D-aspartic acid (NMDA), quisqualic acid (QUIS or AMPA), and kainic acid (KA). Introduction of substituents at the 5-, 7-, and 5,7-positions resulted in analogues having selective NMDA antagonist action, as a result of blockade of the glycine modulatory (or coagonist) site on the NMDA receptor. Regression analysis suggested a requirement for optimally sized, hydrophobic 5- and 7-substituents, with bulk tolerance being greater at the 5-position. Optimization led to the 5-iodo-7-chloro derivative (53), which is the most potent and selective glycine/NMDA antagonist to date (IC50 vs [3H]glycine binding, 32 nM; IC50's for other excitatory amino acid receptor sites, greater than 100 microM). Substitution of 1 at the 6-position resulted in compounds having selective non-NMDA antagonism and 8-substituted compounds were inactive at all receptors. The retention of glycine/NMDA antagonist activity in heterocyclic ring modified analogues, such as the oxanilide 69 and the 2-carboxybenzimidazole 70, suggests that the 4-oxo tautomer of 1 and its derivatives is required for activity. Structurally related quinoxaline-2,3-diones are also glycine/NMDA antagonists, but are not selective and are less potent than the 1 derivatives, and additionally show different structure-activity requirements for aromatic ring substitution. On the basis of these results, a model accounting for glycine receptor binding of the 1 derived antagonists is proposed, comprising (a) size-limited, hydrophobic binding of the benzene ring, (b) hydrogen-bond acceptance by the 4-oxo group, (c) hydrogen-bond donation by the 1-amino group, and (d) a Coulombic attraction of the 2-carboxylate. The model can also account for the binding of quinoxaline-2,3-diones, quinoxalic acids, and 2-carboxybenzimidazoles.