This study investigates two alternative ways in which languages resolve sequences of adjacent vowels (hiatus): deletion of one of the vowels, or coalescence of the adjacent vowels to form a third vowel that combines features of both the originals. Although existing phonological theories predict relatively few restrictions on the behavior of either process, a survey of 92 languages reveals a number of surprising and previously unreported limitations on their behavior. For example, although deletion of the first of two vowels is extremely common and can apply in any position, deletion of the second vowel is restricted to certain well-defined morpho-syntactic contexts, such as the boundary between a root and a suffix. These restrictions, are explained in terms of functionally-motivated constraints that favor preservation of phonological material in certain prominent positions, such as in root morphemes. In the case of coalescence, the study reveals a surprising correlation between the structure of a language's vowel inventory and the result of merging high and a non-high vowels. This correlation is explained in terms of a novel theory of acoustic height features whose detailed specification is determined by functionally-motivated constraints sensitive to the number of vowel heights within a particular language.