Rats were trained to detect an intraperitoneal (IP) administration of cocaine, 10.0 mg/kg, using a two-lever choice discrimination procedure in which food reinforcement was only delivered following 10 responses on the correct lever (FR10): one lever was correct after cocaine injection, and the other lever was correct after saline injection. Following training, cocaine was generalized to the cocaine training stimulus in a dose-dependent manner. Subsequently, guide cannulae were implanted bilaterally in the prefrontal cortex (from bregma, A = 2.7, L = 1.0, V = 3.0 mm), nucleus accumbens (A = 2.2, L = 1.5, V = 6 mm) or caudate putamen (A = 0.2, L = 2.5, V = 4). Injections were made via cannulae that extended 1 mm past the tip of the guide cannulae. Injection in the nucleus accumbens substituted for the IP training dose of cocaine in a dose-dependent manner with maximum generalization occurring with 10 micrograms of cocaine per side (87% cocaine-lever responding); in contrast, injections of cocaine in the prefrontal cortex or caudate-putamen produced only partial cocaine-lever responding (a maximum of 48 and 37% cocaine-lever responding, respectively). These data support the hypothesis of central mediation of the cocaine stimulus and show that cocaine administered in the nucleus accumbens is sufficient to produce the stimulus. The partial substitution of cocaine in the prefrontal cortex and caudate-putamen may reflect partial mediation of the cocaine stimulus in these brain areas.