Theories of phonetic category learning can generally be divided into two camps: rule-based and exemplar-based theories. My previous research has indicated that English speakers trained to learn a phonetic continuum taken from German fricatives violated some of the key predictions of exemplar-based theories. One particularly interesting finding was that listeners tended to find learning discontinuous categories very challenging, sometimes stubbornly responding that each part of the discontinuous category in question was a separate category. Some participants, though, learned the discontinuous categories without much apparent problem. Here, we aim to probe the presentations of the categories that participants learned using a mis-match negativity (MMN) design in MEG. Our prediction is that listeners are merely learning to respond to two underlyingly different categories with the same response, and, thus, will show just as large of an MMN due to category membership for items within the same discontinuous response category as across a true category boundary.