The lexical bias effect refers to the fact that phonological errors result in real words more often than would be predicted by chance. It has also been observed that phonemes are more likely to be exchanged if they are phonologically similar. Both of these patterns of errors are easily explained within the framework of a feedback model (e.g. Dell, 1986), through feedback from phonemes to lexemes, and through feedback from features to phonemes. However, a feed forward account of these effects has also been proposed, which relies on a monitor mechanism that edits out non words and is less likely to reject segments similar to the intended utterance (Nooteboom, 2005) Closer analysis reveals, however, that these tow models make differing predictions concerning the interaction of phonological similarity with the lexical bias effect. The feedback model predicts that connectivity between the feedback loops concerned will result in mutual amplification of the two effects. Therefore, according to the feedback model, lexical bias will increase with phonological similarity. Conversely, Nooteboom (2005) postulates that errors resulting in real words are accepted as lexical by the self-monitor regardless of phonetic similarity, but nonword errors are less likely to be detected if they are phonetically similar to the intended utterance. The adapted monitor model therefore predicts that lexical bias will decrease with increasing phonological similarity This dissertation reports a speech error experiment using the Word Order Competition paradigm (Baars & Motley, 1976), in which phonological similarity and the lexicality of error outcomes are explicitly manipulated. The experiment replicates the lexical bias and phonological similarity effects, previously uninvestigated in this experimental paradigm, and in the interaction uncovered between the two effects adds to the ever increasing pool of evidence for the existence of feedback in language production