Learning to inhibit drug seeking can be an important strategy for inhibiting relapse, and this can be modeled by extinguishing drug seeking in response to a drug-paired context. Rats were either extinguished or withdrawn without extinction training (abstinence) from cocaine self-administration, and measurements of postsynaptic density proteins in the core and shell subcompartments of the nucleus accumbens were compared with yoked-saline controls. Only extinguished rats had elevations of PSD-95, Homer1b/c, and Narp in the postsynaptic density of the core, whereas no proteins measured were altered in the postsynaptic density of the shell in either extinguished or abstinent rats. Using a biotinylation strategy, it was found that surface expression of mGluR5 was reduced only in the core of extinguished animals. Although both extinguished and abstinent animals showed a reduction in long-term potentiation elicited in the core by stimulating prefrontal cortex, blunted long-term depression was observed only in extinguished rats. These data indicate that the elevation in Homer1b/c in the core may have sequestered mGluR5 away from the membrane surface and that the loss of surface mGluR5 inhibits long-term depression. Accordingly, when Homer1c was overexpressed in the core of cocaine-naive rats with an adenoassociated virus, long-term depression was inhibited. This mechanism may contribute to the inhibition of cocaine seeking by extinction training because overexpression of Homer1c in the core also inhibited cue-induced reinstatement of cocaine seeking. These data identify a cellular mechanism that may contribute to extinction-induced inhibition of cocaine seeking.