Assuming that it needs to be decided at some point whether a given Merge(α,β) operation is legitimate, there are two basic options. The first possibility is that one of the two categories is equipped with an intrinsic formal property (typically encoded as a feature) requiring the other one to combine with it. The second possibility is that Merge applies freely throughout, and that filters check the output representation and decide about the legitimacy of the operation. The two approaches are often extensionally equivalent. In this paper, I provide an argument for the first view that is based on the hypothesis that in addition to the Merge operation that builds structure, there is also a mirror image operation Remove that removes structure: If such an operation exists, the legitimacy of the original Merge operation cannot be checked by output filters anymore. Empirical evidence for an elementary syntactic operation Remove is drawn from four domains of German syntax: passive, applicative, restructuring, and complex prefields.