How are the sounds of words represented in plans for speech production? In Experiment 1, subjects produced sequences of four CVCs as many times as possible in 8s. We varied the number of repetitions of the initial consonant, vowel, final consonant, CV, rhyme, and whole CVC each sequence required, and measured subjects' speaking rate. Subjects produced more CVCs when the final consonant or whole word was repeated, but were slowed when only initial sounds or CVs were repeated. Two other experiments replicate the location-based effects and extended them to bisyllabic words. We attribute the locational effects to competition between words that are formally similar, and specifically, to competition between discrepant phonemes in the two words to occupy a particular wordframe position. The fact that only discrepant initial, but not final sounds slow production suggests that phonemes are activated sequentially, from left to right.