Aspect-Oriented Modeling techniques make it possible to use model transformation to achieve advanced separation of concerns within models. Applying aspects that introduce model elements into a base model in the context of large, potentially composite models is nevertheless tricky: when a pointcut model matches several join points within the base model, it is not clear whether the introduced element should be instantiated once for each match, once within each composite, once for the whole model, or based on a more elaborate criteria. This paper argues that in order to enable a modeler to write semantically correct aspects for large, composite models, an aspect weaver must support a flexible instantiation policy for model element introduction. Example models highlighting the need for such a mechanism are shown, and details of how such policies can be implemented are presented.