Using inner eye prostheses, the restoration of vision to the blind has achieved a low level which hopefully will be enhanced in the future. Patients suffering from degeneration of their photoreceptors, cells which modulate light input to neuronal output, can regain visual perceptions by electrically stimulating the remaining retinal neurons. The aim of the investigated modeling approach was to contribute to the understanding of the responses of extracellularly stimulated bipolar and ganglion cells, the primary target cells of current retinal implants. Related to physiological vision, subretinal implants should primarily cause graded potentials in bipolar cells, whereas higher stimuli are needed to directly excite ganglion cells as they are more distant to the electrodes. Epiretinal implants stimulate the retina from the inner portion of the retina with the sensitively excitable axons of ganglion cells closest to the electrodes. Plenty of stimulus-response phenomena, depending on stimulus strength, polarity, cell geometry, ion channel types and other geometric and electrical parameters were systematically investigated. The neural response was calculated in a two-step procedure: i) the electrical field was either obtained with the finite element method or with a simpler analytical approach for point sources and ii) the response of a model neuron was computed by employing a multi-compartment model with the applied electric field as input parameter. Additionally, a model for neurotransmitter release from ribbon synapses at bipolar cell terminals was developed in order to study the temporal impact of L-type calcium channels. Membrane polarization was shown to be stronger for ON than for OFF bipolar cells because of their longer axonal processes. Depolarization of synaptic terminals and consequent vesicle release was only triggered by anodal subretinal stimulation. Strong depolarization above the Nernst potential of calcium, however, led to reversed calcium currents in synaptic terminals. These outward currents prevented an increase of intracellular calcium concentration and consequently less or no neurotransmitter were released. Surprisingly, by stimulating multiple bipolar cells located within a region of 100x100m the calcium reversal led to a pronounced center-surround effect of vesicle release. That is, three stimulation regimes could be discriminated: i) stimulation at low amplitudes did not activate bipolar cells at all (lower threshold), ii) stimulation at appropriate amplitudes only activated bipolar cells close to the stimulating electrode and iii) stimulation in the current reversal regime shut down cells located near the electrode but activated distant cells (upper threshold). The major goal for spatial visual performance is to activate ganglion cells focally, i.e. within a closely spaced region on the retina. Membrane-specific properties such as a distinct distribution of sodium channels of different opening sensitivity (Nav1.2 & Nav1.6) were included into the ganglion cell model. During epiretinal cathodic stimulation, passing axons had thresholds approximately 120% higher than lowest thresholds at the proximal portion of the ganglion cell axon. Consequently, the arising operating window could be used to focally activate a number of ganglion cells without co-activating passing axons from ganglion cells located far away. Generally, thresholds were lower during cathodic stimulation and the site of spike initiation was easier to predict. Anodic stimulation, on the other hand, resulted in complicated activation patterns which hindered to derive general rules for determination of the site of spike initiation. Additionally, simulations suggest that the dendritic portion of the target ganglion cell is also of high importance in spike generation, even when stimulation is applied epiretinally. Dendritic edge compartments (i.e. fiber ends) turned out to have lowered thresholds and therefore played an important role in spike generation under certain circumstances, especially during subretinal stimulation. Furthermore, spike latency was shown to reliably act as a good predictor for site of spike initiation. Adding a noisy transmembrane current component allowed to compute spiking probability as a function of stimulus amplitude resulting in sigmoid response curves similar to experimental determined data. In sum, the spatial and temporal response of retinal neurons was monitored during electrical stimulation with a special emphasis on neurotransmitter release in bipolar cells and the site of spike initiation in ganglion cells. Some of the conclusions could even be found using extremely simplified model neurons, others were confirmed simulating the geometric data of real cells.