I present an analysis of prevocalic initial glottal stops in Amarasi. A range of morphological and phonological evidence shows that a unitary analysis of such glottal stops is not possible: some morphemes occur with an epenthetic initial glottal stop, while other morphemes occur with a contrastive initial glottal stop. This is consistent with the historical data, which show that some glottal stops have developed from an earlier consonant while others are historic insertions. Despite the difference between epenthetic and contrastive glottal stops in Amarasi, there are only a few instances in which the data are unambiguous. In most cases, the status of a particular glottal stop has not yet been determined.