It is usually assumed that the development of speech perception works like this: first, infants use distributional clustering to learn the consonants and vowels of their language; then, young children use these sounds to recognize and distinguish words.  On this account, distributional learning in infancy solves the problem of phonological interpretation in learning words.  I will argue on the contrary that infants may use words to learn speech sounds, and that this learning does not itself cause mature phonological interpretation.  This proposal suggests new ways to consider the interrelation of word learning and phonological learning in infancy.