All the evidence in Sanskrit points to a representation for ambisyllabicity that is no different from that for gemination. Like geminates (on which see Hayes 1986), ambisyllabic segments show ambiguity, behaving as onset segments from the point of view of phonotactics, like coda segments for the purpose of determining weight, and yet with respect to pronunciation, like a single segment. They also show a certain inalterability: even though an ambisyllabic segment closes a syllable and makes position, it is not subject to coda neutralizations. Secondly, they appear to slip in and out of a pronunciation where they are pronounced long, as true geminates. This variation, which is both synchronic and diachronic, is easiest to account for if it is a matter of two noncontrastive phonetic realizations of the same structure. Most importantly, this ambisyllabicity and gemination never contrast, even though the language has both phenomena. Sanskrit contrasts long and short consonants between vowels, as in the aforementioned near-minimal pair vr.'kas vs. vr.kkás. But this contrast obtains only in the case of a single segment, whereas ambisyllabicity in Sanskrit obtains only in the case of consonant clusters, so the difference between the two cases does not require any further structural distinction. The fact that ambisyllabicity and gemination have distinctive environments even when both occur in the same language bolsters the claim that they have identical representation. As for what that is, exactly, my inclination is to conceive of this as a segment being linked into the rime of one syllable and also into the onset of the next, but I do not believe that the data contain any crucial evidence that argue against other versions of autosegmental phonological theory, such as the intermediation of a tier of timing slots, or taking moras to be the top-level constituents of the syllable. What is crucial is that a single segment be linked simultaneously into two adjacent syllables.