The mechanisms underlying the poorer serial recall of talker-variable lists (e.g., alternating female-male voices) as compared with single-voice lists were examined. We tested the novel hypothesis that this "talker variability effect" arises from the tendency for perceptual organization to partition the list into streams based on voice such that the representation of order maps poorly onto the formation of a gestural sequence-output plan assembled in support of the reproduction of the true temporal order of the items. In line with the hypothesis, (a) the presence of a spoken lead-in designed to further promote by-voice perceptual partitioning accentuates the effect (Experiments 1 and 2); (b) the impairment is larger the greater the acoustic coherence is between nonadjacent items: Alternating-voice lists are more poorly recalled than four-voice lists (Experiment 3); and (c) talker variability combines nonadditively with phonological similarity, consistent with the view that both variables disrupt sequence output planning (Experiment 4). The results support the view that serial short-term memory performance reflects the action of sequencing processes embodied within general-purpose perceptual input-processing and gestural output-planning systems. (Contains 2 tables, 4 figures and 5 footnotes.)