In this paper, I propose that blend words can be treated as a form of correspondence between multiple inputs. Based on this suggestion, the theoretical claim underlying my analysis is that prosodic compression is the result of a large prosodic domain being recycled at a lower level of the phonology. The phenomena of phonemic overlap and a preference for sress alignment, which are two of blending's most salient characteristics, fall out from *CLASH, MAX-SEG, and IDENT-STRESS (McCarthy 2008) working together to make the best prosodic word out of the available feet.