Results from an experimental study in Italian show that extraction from prepositional adjuncts has different effects depending on the adjunct taken into account. This can be seen not only from unacceptable sentences containing an island violation, but also from parasitic gaps, which are instead expected to be equally acceptable. In both cases, extraction from the adjunct introduced by senza (without) is more acceptable than extraction from those introduced by dopo (after) and prima (before), which is either completely ruled out in the case of islands, or degraded in parasitic gaps. Moreover, variability within participants is found only in the adjunct introduced by senza. To explain this asymmetry I propose that adjuncts are not a uniform class, and that one of the criteria able to account for their differences is the level of attachment to the structure. Their point of merging is therefore also responsible for the island effect and the opacification of the domain, as well as the possibility of forming a macro-event made by the matrix verb and the one in the adjunct. This latter condition is also required to void the islandhood of adjuncts.