In software product line engineering, feature models enable to automate the generation of productspecific models in conjunction with domain “base models” (e.g. UML models). Two approaches exist: pruning of a large domain model, or merging of model fragments. In this paper, we investigate the impact of the merging approach on base models, and how they are made and used. We adopt an empirical method and test the approach on an example. The results show several challenges in the way model fragments are written, the need for new modelling language constructs and tool support.