1. N-methyl-D-aspartic acid (NMDA)-activated currents were recorded from dissociated rat retinal ganglion cells using whole-cell recording. The NMDA open-channel blocking drug memantine was evaluated for non-competitive and/or uncompetitive components of antagonism. A rapid superfusion system was used to apply various drugs for kinetic analysis. 2. Dose-response data revealed that memantine blocked 200 microM NMDA-evoked responses with a 50% inhibition constant (IC50) of approximately 1 microM at -60 mV and an empirical Hill coefficient of approximately 1. The antagonism followed a bimolecular reaction process. This 1:1 stoichiometry is supported by the fact that the macroscopic blocking rate of memantine (kon) increased linearly with memantine concentration and the macroscopic unblocking rate (koff) was independent of it. The estimated pseudo-first order rate constant for macroscopic blockade was 4 x 10(5) M-1 S-1 and the rate constant for unblocking was 0.44 s-1. Both the blocking and unblocking actions of memantine were well fitted by a single exponential process. 3. The kon for 2 microM memantine decreased with decreasing concentrations of NMDA. By analysing kon behaviour, we estimate that memantine has minimal interaction with the closed-unliganded state of the channel. As channel open probability (Po) approached zero, a small residual action of memantine may be explained by the presence of endogenous glutamate and glycine. 4. Memantine could be trapped within the NMDA-gated channel if it was suddenly closed by fast washout of agonist. The measured gating process of channel activation and deactivation appeared at least 10-20-fold faster than the kinetics of memantine action. By combining the agonist and voltage dependence of antagonism, a trapping scheme was established for further kinetic analysis. 5. With low agonist concentrations, NMDA-gated channels recovered slowly from memantine blockade. By analysing the probability of a channel remaining blocked, we found that memantine binding appeared to stabilize the open conformation of the blocked channel and did not affect ligand affinity. Validity of the "trapping model' and stabilization of the open conformation were further suggested by agreement between the predicted dose-response curve for NMDA in the presence of 2 microM memantine and the empirically derived dose-response relationship. 6. Based on simple molecular schemes, the degree of blockade at various concentrations of agonist for "pure' non-competitive vs. uncompetitive inhibition was computer simulated. The measured degree of blockade by 6 microM memantine was close to ideal for pure uncompetitive antagonism. Taken together, we conclude that the predominant mechanism of open-channel blockade by memantine is uncompetitive. In general, the relative magnitude of the dissociation rate of an open-channel blocker from the open but blocked channel (the apparent off-rate) compared with the rate of leaving the closed and blocker-trapped state (the leak rate) will determine the contribution of uncompetitive vs. non-competitive actions, respectively. 7. Millimolar internal Cs+ competed with memantine for binding in the NMDA-gated channel, and reduced the association rate of memantine, but had no effect on the voltage dependence of the dissociation rate. After removal of Cs+, the calculated Ki for memantine remained voltage dependent. These observations would be difficult to reconcile with models in which memantine binds to a site outside the channel pore and instead strongly support the supposition that the blocking site for memantine is within the permeation pathway.