Spoken words vary considerably in their duration, for a variety of reasons. The current work focused specifically on two of those reasons – the difficulty of lexical retrieval and phonological retrieval and integration – to assess whether they play a significant role in influence the duration of the time to begin speaking and the duration of the words in a simple utterance. Previous work of mine and others has shown that word duration is sensitive to factors that may influence lexical and/or phonological retrieval (Bell et al., 2009; Gahl, 2008; Kahn & Arnold, 2012; Schriefers et al., 1999; inter alia), but never in a context that examined the timing of the mechanisms underlying those factors. The current experiments manipulate the semantic and phonological relationship between a prime and a target utterance in order to investigate the mechanisms and their timing. They also manipulate the difficulty of processing other words in a multi-word utterance than the target, to assess whether word duration is affected by that difficulty. The first two experiments focused on the semantic relationship between prime and target, and found that, consistent with prior work, semantic relatedness between prime and target lengthened utterance latencies to begin speaking, while primes that were identical to the target shortened them. Semantic relatedness did not have a great deal of influence on word duration, however, as contrasted with identical primes (which shortened both latencies and durations) and phonologically-related primes. The second two experiments focused on the phonological relationship between prime and target. Once again consistent with prior work, this relationship elicited shorter latencies. Importantly, it also elicited durational lengthening, most likely as a function of the speaker’s language production system having to ensure that it produces the target’s phonemes instead of the prime’s. Experiments 2 and 4 also included a manipulation of the final word in the utterance (not the target), whether it was consistent between trials or randomly chosen. When it was consistent, it tended to shorten latency to begin speaking, and somewhat less-reliably the duration of the object word and the final word. The findings together show that lexical retrieval is less involved in the determination of duration than phonological retrieval and/or integration. Further, they show that the language production system does not determine the duration of the words in an utt...