The requirement of identifiability for Chinese topics has always been discussed under the framework of reference and definiteness, and limited to nominals, This article argues that this framework is developed against the background of western languages and may not fit the Chinese language properly; and the requirement of identifiability for Chinese topics applies to non-nominals as well. According to general understanding, the identifiability of a nominal (or more precisely, the entity the nominal refers to ) depends on whether the receiver can be assumed to know or to single out from among others the particular one in question. This article demonstrates that this is not the true criterion. Identifiability in fact only depends on whether the receiver can be assumed to be able to lock on to the particular one set by the speaker as topic inthe discourse. Moreover, the speaker does not merely assume, in a passive manner, that the receiver can lock on the topic; he can also actively "process" a topic to give it identifiability even though one and the same topic may not refer to anything the receiver can be assumed to single out in reality. The fact that there are restrictions on the structure "yi+classifier+NP" (not limited to NPs) functioning as topics is because the structure is a marked form of unidentifiability; and it has little to do with its actual identifiability.