We propose a model for the abrupt emergence, below temperatures close to the glass transition, of the ultra-fast (GC) steady mode of spherulitic crystal growth in deeply undercooled liquids. We interpret this phenomenon as controlled by the interplay between the generation of stresses by crystallization and their partial release by flow in the surrounding amorphous visco-elastic matrix. Our model is consistent with both the observed ratios (∼ 10 4 ) of fast-to-slow velocities and the fact that fast growth emerges close to the glass transition. It leads us to conclude that the existence of a fast growth regime requires both (i) a high fragility of the glassformer; (ii) the fine sub-structure specific of spherulites. It finally predicts that the transition is hysteretic, thus allowing for an independent experimental test.Upon approaching T g , the rate of crystal growth from deeply supercooled glassformers decreases to very small values -in the 10 −12 m/s range -as appears consistent with the dramatic slowing down of molecular motions. Growth is hence expected to be nearly arrested in the glass phase. This expectation, however, is challenged by a surprising phenomenon identified by Oguni and coworkers in several fragile materials 1,2 and further investigated, more recently, by the groups of Yu and Ediger at Madison 3 . Namely below some temperature T t slightly larger than T g , an ultrafast crystal growth mode (often coined GC, for Glass-Crystal) suddendly emerges. Near T t the ratio of the fast over normal (slow) front velocities, u f and u s , is huge, typically of order 10 4 . Up to now this phenomenon has been observed in 11 one-component molecular glass-formers 3 , which all have high fragility indices, that is, exhibit the so-called decoupling phenomenon 4,5 . Namely, below ∼ 1.2 T g the viscosity η and translational diffusion coefficient D of the SC liquid no longer obey the Stokes-Einstein relation but instead, D ∼ η −ξ , with 0 < ξ < 1. Ediger et al 6 have shown that, as T varies, the growth velocity of the slow (normal) mode scales, not as η −1 , but as D, a result consistent with the idea that crystallization kinetics is controlled by local processes 7 . Moreover, materials exhibiting fast growth share the following common features: molecules are strongly anisotropic (nearly planar); single-crystals grown at small undercoolings are highly facetted -which indicates slow interfacial kinetics; at large undercoolings, crystal growth is spherulitic 8,9 with, in the usual observation domain (spherulite radii ∼ 10's to 100's of microns), a quasi-constant front velocity.To this day, the paradoxical emergence of this fast growth mode near T g largely remains a mystery. A noticeable proposition formulated by Tanaka 10 is that below T g , volume contraction upon crystallization creates in the surrounding glassy matrix a negative pressure, i.e. free-volume; this would increase particle mobility close to the interface, hence accelerate crystallization. Konishi and Tanaka 11 have shown that the values of the velocit...