Over the last 11 years considerable evidence has accumulated indicating the involvement of NMDA receptors in learning processes. This involves both electrophysiological and behavioural studies using mainly pharmacological approaches. In turn, it is generally accepted that the blockade of NMDA receptors leads to learning impairment while their stimulation, or positive modulation may produce enhancement of learning. However, depending on the model used, exactly the opposite effect can also be seen. We observed that in rats after entorhinal cortex lesions, the uncompetitive NMDA receptor antagonist memantine decreased the frequency of reference memory errors in the radial maze paradigm, but the glycine site partial agonist d-cycloserine was without effect. In contrast, in the passive avoidance test, d-cycloserine, but not memantine attenuated anterograde amnesia produced by scopolamine. Also in the same paradigm, NMDA itself (25–100 mg/kg) induced amnesia that was attenuated by MK-801 and memantine. Similarly, in hippocampal slices 10 μM NMDA depressed AMPA receptor-mediated fEPSPs in the CA1 region and also caused a moderate reduction of LTP induction/expression and this later effect was attenuated by the NMDA receptor antagonist memantine. Moreover, although systemically-active antagonists selective for the glycine site of the NMDA receptors produced an apparent impairment in a passive avoidance test they had no negative effects on radial maze learning. Finally, in view of the essential role of AMPA receptors in glutamatergic transmission, the fact that the AMPA receptor antagonists, NBQX and GYKI-52466 have no effect on learning in a number of paradigms (e.g. passive avoidance, T-maze, radial maze) is very surprising. This diversity of effects indicates that a delicate, physiological balance of AMPA and NMDA receptor activation is a prerequisite for optimal functioning of learning processes and that the effect of NMDA receptor antagonists depends on the test used.