The semicontinuous acoustic signal that we call speech is the result of a number of discrete neuromuscular events. We cannot store in our brain motor commands for every utterance that we may wish to say, owing to the brain's finite storage capacity. Speech then is produced by the rearrangement of a limited number of stored items. What is the size of nature of these stored units? How do speakers encode a sequence of discrete linguistic units into a continuously changing speech signal? An electromyographic study revealed that no simple correspondence exists between a phoneme and its motor commands. Initial and final allophones of consonants differed; both were unaffected by the adjacent vowels. Vowels were influenced by the preceding consonant. The results of this study relevant to the relationship between motor commands and phonetic segments are discussed. Alternative hypotheses relating to the nature of the stored unit are suggested.