In her paper, Arvaniti supports a producton and perception model which distinguishes phonological representations from their phonetic surface form. She draws on intonational data as evidence that an abstract phonological representation is necessary to capture native speakers’competence to extract unique significance when faced with multiple phonetic variations. This view is in direct contrast to models which advocate that contours are holistic and all variation must be coded in detail. My commentary professes sympathy with Arvaniti’s approach and adds further evidence from segmental variation in support of abstract representations.