The amino-adamantane derivatives memantine (l-amino-3,5-dimethyladamantane) and amantadine (1-amino-adamantane) are relatively low affinity, uncompetitive N-methyl-D-aspartate (NMDA) receptor antagonists which have been used clinically in the treatment of dementia and Parkinson's disease respectively for several years without serious side effects. The aim of this study was to test whether memantine, amantadine and other low affinity uncompetitive NMDA receptor antagonists also have better therapeutic indices than high affinity antagonists in preclinical models of epilepsy by assessing the potency, kinetics and voltage-dependency of open channel blockade for a series antagonists in vitro and comparing these effects to anticonvulsive and motor impairment activity in vivo. The compounds tested were memantine, amantadine, 14 other amino-adamantanes, (+)-MK-801, ketamine, dextrorphan, dextromethorphan and phencyclidine. The offset kinetics of open-channel blockade assessed with whole cell patch clamp recordings from cultured superior colliculus neurones were highly correlated to potency i.e. the less potent antagonists showed faster unblocking kinetics (K_off, r = 0.904). Although, onset kinetics as assessed by K_on were not correlated to potency (r = 0.023), τ_on estimated at IC₅₀ is perhaps a more meaningful measure of onset kinetics at equieffective concentrations and was also well correlated to potency (r = − 0.863). All amino-adamantanes tested were strongly voltage-dependent. There was also a good correlation between the in vitro potencies of uncompetitive NMDA receptor antagonists assessed with patch clamp recordings and displacement of equilibrium [³H](+)-MK-801 binding and their in vivo activity against maximal ehxtroshock (MES) and pentylenetetrazol (PTZ) induced tonic convulsions and NMDA-induced lethality in mice. Memantine and four other amino-adamantanes with somewhat lower potency and faster blocking kinetics had better therapeutic indices (ED₅₀ rotarod and traction reflex over ED₅₀ in MES-induced convulsions; TI = 2–4) than substances with higher affinity such as ketamine, dextrorphan and (+)-MK-801 (TI < 2). However, amantadine and several other amino-adamantanes with lower potency than memantine actually had poorer therapeutic indices (TI ⩽ 0.5) which may have been due to additional actions at other ion channels or receptors at the doses necessary to protect against seizures. In fact, ED⁵⁰ in the MES test was negatively-correlated to therapeutic indices (traction r = − 0.790, rotarod r = − 0.797) i.e. the less potent uncompetitive antagonists had worse therapeutic indices. The data from the present study do not lend support to the idea that low affinity, open channel NMDA receptor blockers are also effective in models of epilepsy at doses having little effect on physiological processes. It should be stressed that these data do not contradict the known therapeutic safety of memantine and amantadine in dementia and Parkinson's disease respectively. Thus the good clinical profile of memantine in dementia has been attributed not only to its fast blocking/unblocking kinetics but also to its strong voltage-dependency. These biophysical properties may allow therapeutically-relevant concentrations to block chronic, low level pathological activation of NMDA receptors whilst leaving their synaptic activation intact. Precisely these properties may also underlie the poor therapeutic indices seen in the present study on antiepileptic activity due to the synaptic nature of both seizures and normal glutamatergic transmission.