The purpose of the current work is the application of a recent extension (Reusch et al., 2003a, 2003b) of the Gurson-Needleman-Tvergaard (GTN) model (e.g., Needleman and Tvergaard, 1984) to the simulation of ductile damage and failure processes in metal matrix composites at the microstructural level. The extended model is based on the treatment of void coalescence as a lengthscale-dependent process. In particular, we compare the predictions of the (local) with GTN model with those of the lengthscale-dependent extension for ductile crack initiation in ideal and real Al-SiC metal matrix microstructures. As shown by the current results for metal matrix composites and as expected, the simulation results based on the local GTN model for both the structural response and predicted crack path at the microstructural level in metal matrix composites are strongly mesh-dependent. On the other hand, those based on the current lengthscale-dependent void-coalescence modelling approach are mesh-independent. This correlates with the fact that, in contrast to the local approach, the predictions of the lengthscale-dependent approach for the crack propagation path in the real Al-SiC metal matrix composite microstucture considered here agree well with the experimentally-determined path.