There is a significant body of psycholinguistic evidence that supports the hypothesis of an access to constituent representation during the mental processing of compound words. However it is not clear whether the internal hierarchy of the constituents (i.e., headedness) plays a role in their mental lexical processing and it is not possible to disentangle the effect of headedness from that of constituent position in languages that admit only head-final compounds, like English or Dutch. The present study addresses this issue in two constituent priming experiments (SOA 300ms) with a lexical decision task. Italian endocentric (head-initial and head-final) and exocentric nominal compounds were employed as stimuli and the position of the primed constituent was manipulated. A first-level priming effect was found, confirming the automatic access to constituent representation. Moreover, in head-final compounds data reveal a larger priming effect for the head than for the modifying constituent. These results suggest that different kinds of compounds have a different representation at mental level: while head-final compounds are represented with an internal head-modifier hierarchy, head-initial and exocentric compounds have a lexicalised, internally flat representation.