The creation of sentences has been argued to proceed from a syntactic to a phonetic plan, with no interaction between the two. Three experiments investigated a phenomenon that appears to contradict the hypothesized absence of interaction. Coupling phonological priming with a sentence production task, Experiments 1 and 3 showed that words with accessible forms tended to precede those with less accessible forms. This occurred when word order changes required syntactic and morphological changes as well as when they did not. Experiment 2 found the same pattern when words were produced in lists rather than in sentences, further implicating a lexical source. Differences in accessibility also affected hesitations and false starts, suggesting that dysfluency and structural variation sometimes reflect the same underlying problem, a disruption in coordinating word forms with syntactic plans.