The present experiment evaluates whether irregular words show a processing advantage in a reading aloud task, when they have a homophonic partner, which is spelled regularly (e.g., ‘suite’ vs. ‘sweet’). In order to compare the homophonic reading performance to a control condition, irregular nonhomophonic words are also given for reading aloud (e.g., ‘swan’). According to Levelt, Roelofs, & Meyer (1999) a processin advantage for homophones can be expected because their model postulates a shared entry on word form level. The results show that irregular homophones do not benefit from the postulated shared entry in the phonological output lexicon. We find a homophone disadvantage in the reading aloud task: Homophones are read slower than their nonhomophone controls. We investigate further whether the Dual Route Cascaded (DRC) model of reading can simulate our results comparing two different versions of the model: one with a single phonological representation for homophones and a second version with two separate representations for homophones.