The L-grass model is a functional-structural model simulating the morphogenesis of perennial grasses above- and belowground. However, due to the high plant density in grasslands and the great number of phytomers and organs to consider over the years, its running time follows a sub-exponential function of the number of iterations. The objective of this study was to build a meta-model of the tiller architectural characteristics (namely the rate of phytomer production and the final dimensions of leaves) able to account for a wide range of genotypes, light competition regimes and management practices (i.e. response to defoliation), while presenting reduced simulation time. Two meta-modelling strategies were compared. The first consisted in "direct" empirical relationships between input parameters and output variables. The second consisted in building a series of "nested" relationships based on intermediate variables that mimicked the L-grass functioning to regulate grass morphogenesis. Our results showed that both strategies were able to accurately reproduce L-grass simulations for independent series of datasets in absence of defoliation (i.e. genotype and light competition effects). However, only the "nested" meta-modelling approach was able to account for the plastic plant response induced by defoliation in terms of leaf size and phyllochron.