To study the role of glutamate in cocaine-conditioned responses, we developed a rat model in which conditioned locomotion is produced by repeated pairing of cocaine with discrete stimuli (flashing light and metronome). ‘Paired’ subjects received cocaine (15 mg/kg) prior to six exposures to stimuli for 30 min in the test environment. ‘Unpaired’ subjects received equivalent presentations of the stimuli yet received cocaine in home cages. Tests with the stimuli alone demonstrated that the conditioned locomotion displayed by Paired subjects was evident at 3 or 10 days post-training and resistant to two sessions of testing. The degree of conditioned locomotion was not correlated with the subjects' response to novelty or cocaine. Administration of the noncompetitive AMPA receptor antagonist GYKI 52466 (2.5 mg/kg, a dose without effect on spontaneous activity) attenuated the expression of conditioned activity. In vivo microdialysis revealed that Paired subjects had significantly lower basal glutamate levels in the nucleus accumbens (NAc) than did Unpaired subjects when no stimuli were presented. Presentation of the conditioned stimuli resulted in significant increases in glutamate levels in the NAc in the Paired group whilst glutamate levels in the Unpaired group remained unchanged. The associative control of glutamate levels in the NAc by stimuli formerly paired with a drug of abuse is an unprecedented finding. It is likely to reflect the convergence of excitatory inputs that the NAc receives from limbic structures.