Superelasticity (SE) in bulk materials is known to originate from the structure-changing martensitic transition which provides a volumetric thermodynamic driving force for shape recovery. On the other hand, structure-invariant deformation processes, such as twinning and dislocation slip, which result in plastic deformation, cannot provide the driving force for shape recovery. We use molecular-dynamics simulations to show that some bcc metal nanowires exhibit SE by a “reversible” twinning mechanism, in contrast to the above conventional point of view. We show that this reversible twinning is driven by the surface energy change between the twinned and detwinned state. In view of similar recent findings in fcc nanowires, we suggest that SE is a general phenomenon in cubic nanowires and that the driving force for the shape recovery arises from minimizing the surface energy. Furthermore, we find that SE in bcc nanowires is unique in several respects: first, the ‹111› / {112} stacking fault generated by partial dislocation is always preferred over ‹111› / {110} and ‹111› /{123} full dislocation slip. The occurrence of ‹111› / {112} twin or full dislocation slip in bcc nanowires depends on the competition between the emission of subsequent partial dislocations in adjacent {112} planes and the emission of partial dislocations in the same plane. Second, compared to their fcc counterparts, bcc nanowires have a higher energy barrier for the nucleation of twins, but a lower energy barrier for twin migration. This results in certain unique characteristics of SE in bcc nanowires, such as low energy dissipation and low strain hardening. Third, certain refractory bcc nanowires, such as W and Mo, can show SE at very high temperatures, which are higher than almost all of the reported high-temperature shape memory alloys. Our work provides a deeper understanding of superelasticity in nanowires and refractory bcc nanowires are potential candidates for applications in nanoelectromechanical systems operating over a wide temperature range.