Linguists are in general agreement that phonemes can be recognised by their ability to create lexical contrast. The method by which phonemic status is established involves finding word pairs that differ on a single sound unit. If two phones [X] and [Y] can be shown to function contrastively, they are regarded as separate phonemes; if not, they are considered allophones of the same underlying unit. A central thesis of this paper is that this defining aspect of the phoneme is highly problematical and should be abandoned. In particular, the empirically testable consequence of the phoneme, functional load, is shown to be confounded with frequency of occurrence and to be unable to account for the order of acquisition of English consonants and their production difficulty in aphasia. An alternative psychologically-based view of the phoneme is developed according to which the phoneme is understood as a prototype which induces contrast at the phonological but not at the lexical level. This newly defined unit, termed the segment, is less abstract than the phoneme but none the less capable of dealing with problems which have been believed to require phonemic elements. The segmental account divides into a strong and a weak version. The former allows no effects of lexical contrast at all on phonological processing and representations while the latter is slightly more liberal in this respect. It is tentatively concluded that the weak version stands a better chance of being borne out in the long run than the strong version, given the highly interactive nature of the processing system. © 1993, Walter de Gruyter. All rights reserved.