The underlying form is an important theoretical concept in generative phonology,and also an important part of its grammatical operation.The underlying form of generative phonology is not the structuralist"phoneme";there is no"phoneme"or"morphophonemic level"in generative phonology,as Chomsky and Halle(1968:11)have clearly stated.Due to their misunderstanding,many Chinese scholars often take"phoneme"as the underlying form of generative linguistics.It should be known that the grammar architecture of structural linguistics before the 1950s is a discovery procedure,by which the collected data are processed through the three successive levels of phonemics,morphology and syntax,and finally the grammar of a particular language under discussion is constructed.The phonological analysis of structural linguistics is mainly to find out and set up the phonemic inventory by contrastive analysis of minimal pairs.Generative phonology since the 1950s,however,is a mapping or derivation mechanism from underlying representation(UR)to surface representation(PR).The UR is abstract and is posited in accordance with various phonological conditions and requirements,while the PR is concrete and is the actual phonetic realization.As an important concept of structural phonology,a phoneme is neither concrete nor abstract,which is simply an intermediate stage in the phonological mapping or derivation process.Thus,it is inappropriate to change the"underlying vowels or consonants"used in some of the PhD dissertations into the"phonemes"in Chinese language.The present paper will start with the definition of the structuralist"phoneme",then make the more detailed explanation of differences between the phoneme and the underlying form,pointing out that the underlying form is not a form that must appear on the surface and it may also be a totally hypothetical form.The underspecification theory,in fact,has nothing to do with the way of choosing the underlying forms,and the function of underspecification theory is to assign a certain number of feature values to the underlying forms,rather than to select the underlying forms.Thus,it is completely wrong to take two underspecification theories as a tool for choosing the underlying forms.Therefore,when we conduct phonological analysis within the generative framework,we must be careful not to mistakenly incorporate the abandoned concepts and practices of structuralist linguistics into it,but must keep in mind that A is A and B is B.Only in this way can satisfactory analysis and results be obtained.