The research presented here is concerned with the phonological nature of mental representations of native speakers. Following the FUL-model (Featurally Underspecified Lexicon, Lahiri & Reetz 2002), we claim that not all features are present in the mental representation and that the grammar severely constrains the way in which phonological contrasts are preserved and utilized in normal language use. Evidence for the FUL-model comes from language changes which extend and maintain phonological contrasts. The model is also supported by results of our MEG and EEG studies which demonstrate different cortical structures for acoustically equidistant but phonologically conflicting vowels, and enhanced automatic change detection response in the brain (mismatch negativity) in instances of feature conflict (Obleser, Elbert, Lahiri & Eulitz 2001, Eulitz & Lahiri 2002).