After spinal cord injury (SCI), the communication between the brain and spinal neural circuits below the lesion is disrupted, leading to paralysis. Epidural electrical stimulation (EES) of the lumbosacral spinal cord has been shown to restore locomotion and promote long-term recovery of voluntary leg control in individuals affected by SCI. EES predominantly engages sensory nerve fibers in dorsal roots and dorsal columns, modulating the motor output of lumbosacral spinal segments via mono- and poly-synaptic pathways. Similar sensorimotor circuits involved in upper-limb motor control are located in the cervical segments, suggesting that EES could also be applied to promote arm and hand function in tetraplegic patients after SCI. However, the ability of EES to modulate specific cervical motor nuclei, likely essential to promote upper-limb function, remains largely unexplored. Hybrid neurophysical volume conductor models have been valuable to unravel the mechanisms of lumbosacral EES and are still intensively used to design implantable electrode arrays and stimulation strategies. Via a two-steps computational scheme in which geometrical factors have a significant influence, they allow to compute the tridimensional electric potential induced in biological tissues and assess the electrical response of neurons and nerve fibers. In this thesis, I have conceived a new geometrical model to represent the spinal cord including notably the spinal roots, I refined the geometry of spinal nerve fibers, and I have developed a software suite implementing these models and enabling the semi-automatic construction of detailed hybrid models from user configuration datasets. I used this suite to construct a hybrid model of EES of the macaque monkey cervical spinal cord to explore the potential targets of cervical EES and evaluate the recruitment selectivity of epidural electrode arrays specifically tailored to the cervical spinal cord. Electrophysiological experiments led on actual macaque monkeys overall corroborated the findings from the model: cervical EES can selectively modulate specific upper-limb motor nuclei via the direct recruitment of dorsal root primary afferents, but the specificity is limited notably by the mix of distinct afferent populations in the dorsal roots. One way to increase this specificity might be to target individual rootlets, for which multipolar stimulation configurations could be decisive. I have reviewed the strategies employed to represent and search for optimal multipolar configurations using hybrid models, and I propose a new economic and versatile approach whose validity is proven on the theoretical ground.