Although many nervous system disorders are associated with N-methyl-D-aspartate (NMDA) receptor overactivation, pharmacological inhibition of NMDA receptors has typically demonstrated limited clinical value due to debilitating psychotomimetic side-effects. Memantine, however, induces far fewer behavioural side-effects than other NMDA receptor channel blockers such as ketamine, and slows the progressive cognitive decline associated with Alzheimer's disease. Memantine and ketamine inhibit NMDA receptors with similar affinity and kinetics. A prominent mechanistic difference between memantine and ketamine is the degree to which they are 'trapped' within the closed channel of NMDA receptors following removal of agonist: ketamine becomes trapped in nearly all NMDA receptors to which it was bound before agonist removal, whereas some bound memantine molecules dissociate after agonist removal, a phenomenon called partial trapping. Here we investigated the mechanism underlying partial trapping of memantine by recombinant NR1/2A NMDA receptors. We found that memantine dissociation from NR1/2A receptors after agonist removal (the process that results in partial trapping) followed an exponential time course with tau = 0.79 +/- 0.32 s. Neither membrane voltage depolarization nor maintained presence of memantine after agonist removal affected partial trapping, suggesting that partial trapping does not result from memantine escape through open channels. We tested the hypothesis that partial trapping results from binding of memantine to two sites, a superficial 'non-trapping' site and a deep 'trapping' site, which cannot be occupied simultaneously. This hypothesis was supported by the lack of ketamine binding to the superficial site, the voltage dependence of partial trapping, and the effect on partial trapping of a mutation near the deep site. The superficial binding site for memantine may, by causing partial trapping, contribute to memantine's unique therapeutic utility.