The nucleation of recrystallization in deformed and annealed metals is reviewed. The main mechanisms are thought to involve the growth of subgrains by low angle boundary (LAGB) migration in an orientation gradient or the strain induced boundary migration (SIBM) of existing boundaries. Although these mechanisms are reasonably well understood, the details of the dislocation recovery mechanisms which are often required before migration can occur, particularly in metals in which recovery is slow, are poorly understood. Complete experimental investigation of the nucleation event requires a 3-d in-situ technique which will resolve dislocations, and this is not currently available. Although recrystallized grains of orientations not in the deformed structure have been reported, there is as yet no substantial evidence or theory to suggest the creation of new orientations by mechanisms other than annealing twinning. It is concluded that further understanding of the deformed state is required before adequate models of nucleation can be formulated and verified.