|Abstract:||This dissertation proposes the Splitting theory of consonant epenthesis, which incorporates two key ideas: (1) there is a phonological operation called 'splitting', where an input segment corresponds to multiple output segments, and (2) there is no insertion operation involving consonants. Within Splitting theory, epenthetic consonants always correspond to an input segment, and the mapping is always regulated by constraints requiring input-output identity. From this perspective, homorganic glide epenthesis next to high vowels is the most faithful epenthesis possible. For example, in the mapping /i/->[ji], input /i/ corresponds to both [j] and [i] in the output, and both output segments preserve all input features. Splitting theory predicts that epenthetic consonants will be as faithful to their input vowel correspondent as possible. However, in some cases perfect faithfulness is not achievable. For example, there is no featurally identical glide counterpart of non-high vowels. Consequently, next to non-high vowels there are several options for epenthetic consonants, all of which preserve some of the input's features. If the most faithful option is banned from a surface inventory, the epenthetic consonant has to be unfaithful to its input correspondent. In these cases, the quality of inserted consonants is also affected by the ranking of faithfulness constraints. An extreme case is found in Mongolian, where a dorsal/uvular stop is epenthesized in vowel hiatus because there is no other consonant that would preserve the place of articulation, voicing, and non-nasality of underlying vowels. Splitting theory's emphasis on faithfulness differs from theories where the epenthetic segment does not correspond to any input segment. For example, such theories predict that epenthetic [t] is possible, while Splitting theory imposes very restrictive (and practically insurmountable) conditions on any system having epenthetic [t]. Putative cases of epenthetic [t] are shown to admit alternative analyses (e.g. Ajyíninka Apurucayali). Splitting Theory is supported by an in-depth survey of the inventory of epenthetic consonants. The theory is illustrated by analyses of consonant epenthesis in Dutch, English, Faroese, Madurese, Mongolian, and Washo.