Four experiments reexamined the hypothesis that immediate serial recall is limited by the spoken duration of list items during output. In all the experiments, spoken serial recall was measured for visually presented words of short and long duration. The duration of first- and second-half list items was varied independently. Experiment 1 used the disyllabic words from Lovatt, Avons, and Masterson (2000). No main effect of the duration of words recalled first was obtained on accuracy of recall. Experiment 2 used the disyllabic words from Cowan et al. (1992). For this set of words, placing long items at the beginning of the list brought about a decrease in recall as predicted by theories of output decay. Experiment 3 confirmed that an effect of duration of words recalled first was confined to the Cowan et al. words, but this effect vanished when the first-half errors were statistically controlled or excluded. Experiment 4 extended these findings to the disyllabic words originally selected by Caplan, Rochon, and Waters (1992) and a new set of three-syllable words. For all these word sets, speech time during output differed for the long and short words. However, for two sets there was no effect of first-half word duration, and for the other two sets first-half word duration was confounded with first-half error rate. Taken together, the results provide no evidence that the spoken duration of early list items directly affects later recall, as predicted by the speech time hypothesis.