Dislocation microstructures in olivine grains from experimentally deformed specimens as well as from peridotite xenoliths in lavas indicate that (1) edge dislocations probably control the creep rate and (2) {110} planes exert a strong crystallographic control on b=[100> edge dislocations even at very high temperature (0,9 Tm). This observation can be explained by a sessile splitting or by the existence of a deep Peierls valley in {110} planes. In either case, kink and jog nucleation are likely to be difficult so that creep can be controlled either by kink or jog formation. This interpretation implies that different activation energies should be associated with the different glide systems. It is in agreement with the fact that the activation energy for creep is not that for oxygen diffusion.