[en] Model Driven Engineering (MDE) was identified as a viable software development paradigm to help improve the product derivation phase of the Software Product Line (SPL) engineering process. Existing model-driven derivation approaches fail to properly address the behavioural derivation part, yielding a frustrating situation. In this paper we first introduce a modeldriven derivation approach that combines Feature Diagrams (FD) and model fragments. We then identify and analyse several issues that emerge during the derivation process. We show that the order in which models associated to selected features are composed has a great impact on the end result of the derivation. We also present a particular class of features called disjoint and prove that current composition operators do not offer any viable solution to compose them. Finally, we argue that insufficient information available to composition operators leads to derivation results that do not satisfy user requirements.