In some models of phonology as well as morphophonology, the underlying representation (UR) or underlying form (UF) of a word or morpheme is the abstract form the word or morpheme is postulated to have before any phonological rules have applied to it. If more rules apply to the same form, they can apply wholly independently of each other or in a feeding or counterbleeding order. The underlying representation of a morpheme is considered to be invariable across related forms (except in cases of suppletion), despite alternations among various allophones on the surface.
In many cases, the underlying form is simply the phonemic form. For example, in many varieties of American English the phoneme /t/ in a word like wet can surface either as a glottalized [tˀ] or as a flap [ɾ], depending on environment: [ˈwɛtˀ] wet vs. [ˈwɛɾɚ] wetter. In both cases, however, the underlying representation of the morpheme wet is the same: its phonemic form /wɛt/.
In other cases, phonological rules may change the phonemes involved. In such cases, pipes ("|") or double slashes may be used in transcription to distinguish the underlying form from its phonemic realization. An example is the word cats, which has the phonemic representation /ˈkæts/. If we take the underlying form of the English plural suffix to a 'z' sound, as many phonologists do, then the underlying form would be //ˈkætz//. This discrepancy cannot be avoided by choosing a different underlying form of the plural: If we assume it is an 's' sound, as it's spelled, then the word dogs, phonemically /ˈdɒɡz/, would have the underlying form //ˈdɒɡs// (in both cases, there has been progressive assimilation of the final segment). As this example illustrates, the underlying form of a morpheme is a theoretical construct, and depends on the analysis that is used.
Sandhi, such as the tone sandhi of the Chinese languages, is another phonological process that changes the phonemes of a morpheme from its underlying form.