We discovered in slices of rat visual cortex that reliable long-term potentiation (LTP) of synaptic responses in layer III could be elicited by theta burst stimulation delivered to a site in the middle of the cortical thickness, corresponding mainly to layer IV. This synaptic plasticity was reflected in the extracellular field potentials and intracellular EPSPs in layer III, but was not observed in the intracellular responses of layer V neurons, suggesting a preferential involvement of synapses on layer III neurons. Tetanus-induced LTP in this preparation was input specific, and was blocked by application of an NMDA receptor antagonist (but not by an antagonist of nitric oxide synthase). In addition, LTP of layer IV-evoked responses could also be produced reliably by pairing low-frequency synaptic stimulation (approximately 100 pulses at 1 Hz) with strong intracellular depolarization of layer III neurons. Thus, LTP in this circuit satisfies the definition of a “Hebbian” modification. Tetanic stimulation of the white matter, in sharp contrast, consistently failed to elicit LTP in layer III unless a GABAA receptor antagonist was applied to the slice. Analysis indicated that the critical difference between layer IV and white matter stimulation was not the magnitude of the responses to single stimuli delivered to the two sites, but that it might lie in the postsynaptic response during high-frequency stimulation. Consistent with this idea, “associative” LTP could be elicited from white matter when converging but independent inputs from the white matter and layer IV simultaneously received tetanic conditioning stimulation. A hypothetical model is presented to account for the differences between layer IV and white matter stimulation. According to this “plasticity gate hypothesis,” inhibitory circuitry in layer IV normally acts as a sort of band-pass filter that constrains the types of activity patterns that can gain access to the modifiable synapses in layer III. By stimulating in layer IV, we have bypassed this filter and therefore do not need to block GABAA receptors to achieve the threshold for LTP in layer III.