For the simulation of the thermoforming of semi-crystalline thermoplastics, an accurate description of melting kinetics is crucial as the temperature field strongly affects the stretching behaviour of the polymer in a short range of temperature. Below the melting temperature, the polymer is too stiff because of the crystalline structure and it rapidly looses its plastic behaviour above this temperature. The heat absorbed by the material during the phase change is usually taken into account in the heat balance equation using a kinetic source term, which generally involves an inverse Avrami model for the description of the time-temperature dependency of the phase change. This model only requires few parameters and can be easily implemented. This approach has demonstrated its relevance for the simulation of thermoplastic welding that requires a total polymer fusion. Crystallization models that are based on a nucleation and growth mechanisms approach however fail in properly predicting melting kinetics. This is due to the large dissymmetry of melting endotherms that results from the distribution of crystallite size in the polymer. Based on this assumption, Greco et al. (2002) have demonstrated that statistical models are relevant models for describing melting kinetics. In particular, they showed that melting kinetics could be modelled with a Richards function. The microcrystalline structure is however not necessarily monomodal. For instance stepwise cooling can induce multiple melting peaks. The present study then further investigates the statistical approach in order to propose a melting model providing an accurate prediction of polymer fusion. The Gibbs-Tompson equation is used to determine the lamellae thickness distribution and polymer fusion is described with a limited series of kinetic models according to crystallite distribution. The model validity is analysed for different materials ranging from PE to PEEK previously crystallized under various cooling conditions