Low-Intensity Focused Ultrasound Stimulation (LIFUS) holds promise for the remote modulation of neural activity, but an incomplete mechanistic characterization hinders its clinical maturation. Here we developed a computational framework to model intramembrane cavitation (a candidate mechanism) in multi-compartment, morphologically structured neuron models, and used it to investigate ultrasound neuromodulation of peripheral nerves. We predict that by engaging membrane mechanoelectrical coupling, LIFUS exploits fiber-specific differences in membrane conductance and capacitance to selectively recruit myelinated and/or unmyelinated axons in distinct parametric subspaces, allowing to modulate their activity concurrently and independently over physiologically relevant spiking frequency ranges. These theoretical results consistently explain recent empirical findings and suggest that LIFUS can simultaneously, yet selectively, engage different neural pathways, opening up opportunities for peripheral neuromodulation currently not addressable by electrical stimulation. More generally, our framework is readily applicable to other neural targets to establish application-specific LIFUS protocols.