The vestibular organs are very important to generate reflexes critical for stabilizing gaze and body posture. Vestibular diseases significantly reduce the quality of life of people who are affected by them. Some research groups have recently started developing vestibular neuroprostheses to mitigate these symptoms. However, many scientific and technological issues need to be addressed to optimise their use in clinical trials. We developed a computational model able to mimic the response of human vestibular nerves and which can be exploited for "in-silico" testing of new strategies to design implantable vestibular prostheses. First, a digital model of the vestibular system was reconstructed from anatomical data. Monopolar stimulation was delivered at different positions and distances from ampullary nerves. The electrical potential induced by the injected current was computed through finite-element methods and drove extra-cellular stimulation of fibers in the vestibular, facial, and cochlear nerves. The electrical activity of vestibular nerves and the resulting eye movements elicited by different stimulation protocols were investigated. A set of electrode configurations was analyzed in terms of selectivity at increasing injected current. Electrode position along the nerve plays a major role in producing undesired activity in other nontargeted nerves, whereas distance from the fiber does not significantly affect selectivity. Indications are provided to minimize misalignment in nonoptimal electrode locations. Eye movements elicited by the different stimulation protocols are calculated and compared to experimental values, for the purpose of model validation.