Infants learn to segment words from fluent speech during the same period as they learn the sounds of their native language, yet accounts of phonetic category acquisition typically ignore information about the words in which sounds appear. This talk presents computational and empirical work testing the hypothesis that the words infants segment from fluent speech provide a useful cue for learning about phonetic categories. Computational models illustrate how words can help an ideal learner acquire English vowels, and experiments demonstrate that real learners are sensitive to the distribution of sounds in words. Results support an interactive account of phonetic category acquisition but highlight a number of open questions about the nature of this interactive learning process.