The most common composition mechanism for Petri nets synchronizes transitions of two nets by merging them: two transitions are merged into one whose input places consists of all input places of the two transitions in their respective nets, and accordingly for the output places. This approach seems appropriate in applications where components are permanently connected, and therefore their models can be composed. However, it becomes very complex and even unsuitable when dealing with dynamically changing connection relations. This contribution introduces a different composition mechanism tailored for the dynamic composition of Petri nets. We will call this interaction. The core idea is that when Petri net components interact, corresponding (interacting) transitions are merged, as is the case for composition, but the original transitions are preserved and available for further interactions. The approach is motivated by various Petri net examples and applied to Workflow nets, where it incorporates several early concepts introduced by Wil van der Aalst.