Whole-cell recordings were obtained from retinal ganglion cells of the tiger salamander (Ambystoma tigrinum) in a superfused slice preparation to evaluate contributions of NMDA (N-methyl-D-aspartate) and KA/AMPA (kainate/α-amino-3-hydroxy-5-methyl-4-isoxalone propionic acid) receptors to excitatory postsynaptic potentials (EPSPs) of retinal ganglion cells. Synaptic activation of retinal ganglion cells was achieved through the use of a brief pressure pulse of hyperosmotic Ringer (Ringer + sucrose) delivered through a microelectrode visually placed in the inner plexiform layer while whole-cell recordings were obtained from adjacent cells in the ganglion cell layer. Separation of NMDA and KA/AMPA excitatory postsynaptic currents (EPSCs) was achieved through the application of the antagonists NBQX and D-AP7, while inhibitory currents were blocked by strychnine and picrotoxin. Simple addition of the two independent EPSCs showed, most often, that the sum of the KA/AMPA and NMDA currents was less than the control response, but in some cases the sum of the two currents exceeded the magnitude of the control response. Neither result was consistent with expectations based on voltage-clamp principles and the assumption that the two currents were independent; for this reason, we considered the possibility of nonlinear interactions between KA/AMPA and NMDA receptors. Computer simulations were carried out to evaluate the summation experiments. We used both an equivalent cylinder model and a more realistic, compartmental model of a ganglion cell constrained by a passive leakage conductance, a linear KA/AMPA synaptic current, and a nonlinear NMDA current based on the well-known, voltage-sensitive Mg2+ block. Computer simulation studies suggest that the hypo- and hyper-summation of NMDA and KA/AMPA currents, observed physiologically, can be accounted for by a failure to adequately space clamp the neuron. Clamp failure leads to enhanced NMDA currents as the ion channels are relieved of the Mg2+ block; their contribution is thus exaggerated depending on the magnitude of the conductance change and the spatial location of the synaptic input.