In simulations with artificial neural networks, efficient information processing and storage has been shown to require that the strength of connections between network elements has the capacity to both increase and decrease in a use-dependent manner. In contrast to long-term potentiation (LTP) of excitatory synaptic transmission, activity-dependent long-term depression (LTD) has been difficult to demonstrate in forebrain in vivo. Theoretical arguments indicate that coincidence of presynaptic excitation and low-magnitude postsynaptic activation are the necessary prerequisites for LTD induction. Here we report that stimulation paradigms which cause 1) sufficient excitation to result in NMDA receptor activation and simultaneously 2) attenuate the level of postsynaptic activation by recruitment of GABAA receptor-mediated inhibition consistently produce LTD of commissural input to area CA1 in the hippocampus of anesthetized adult rats, and of the perforant path input to the dentate gyrus in the hippocampus of anesthetized and unanesthetized adult rabbits. A functionally similar pre- and postsynaptic activation pattern applied to the hippocampal slice preparation by injecting hyperpolarizing current into the postsynaptic cell during NMDA receptor-mediated excitation also was effective in consistently inducing LTD. Results of studies in vitro show that Ca2+ influx through the NMDA channel is necessary for the induction of LTD, and moreover, that NMDA receptors also participate in the expression of LTD. Our findings demonstrate a general mechanism for the implementation of a theoretically derived learning rule in adult forebrain in vivo and in vitro and provide justification for the inclusion of use-dependent decreases of connection weights in formal models of cognitive processing.