A model for phase transitions initiated on grain boundaries is proposed and tested against numerical simulations: this approach based on a grain explicit model (GEM) allows to consider the granular structure, yielding accurate predictions for a wide span of nucleation processes. Comparisons are made with classical models of homogeneous (JMAK [1]) as well as heterogeneous (Cahn [2]) nucleation. A transition scale based on material properties is proposed, allowing to discriminate between random and site saturated regimes. Finally, we discuss the relationship between an Avrami type exponent and the transition regime, drawing conditions for its extraction from experiments.PACS numbers: 05.70. Fh, 81.10.Aj, Recrystallization is a mechanism of great scientific and technological importance, encountered during the thermomechanical processing of various materials including metals [3]. The first model that efficiently captures the main features of crystallization, namely the JMAK model, states that grains nucleate from points of random locations, and that the grains grow until impinging other neighboring growing grains. Thanks to its straightforwardness, this model was also used in many other situations, provided that the hypotheses of random nuclei and independent growing zone are met: temperature dependant crystallization [4], combustion [5], particle physics [6], crystallization in amorphous materials [7], evolution of damage under dynamic tensile loadings [8], and solid state phase transitions in general [9], making the JMAK model a much encountered approach.Random distribution of nuclei is however a rather crude hypothesis. To give examples, materials experience damage by crack nucleation preferably at grain boundaries [10,11], combustion of solid energetic materials starts at preferred sites [12], nucleation of grains during recrystallization appears at prior grains frontier [13], microstructuring nanocomposites enhances the kinetics of physisorption [14], and crystallization can be influenced by impurities [15] or confinement in a porous media [16] or contact with grain boundaries of an other material [17]. Thus, the problem of nucleation from interfaces is encountered in a variety of fields [18][19][20][21][22][23][24] and the importance of structured nucleation sites -inner nucleation free volumes bounded by interfaces -as well as grain size dependence is commonly witnessed [2,25].In this regard, extensions of the JMAK model have been proposed over the years to take into account the specificities of structured nucleation. Most derivations still consider random distribution of nuclei, improving only marginally the model by fitting so-called Avrami parameters, which is seen as lacking clear physical justification [26][27][28].One of the major improvements of the JMAK model that faces the problem of nucleation heterogeneity has been proposed by J.W. Cahn and considers nuclei distributed on planar interfaces [2,29]. This approximation extends the predictions to heterogeneous nucleation, provided that the density of ...