Two problems for theorists of speech perception are those of segmentation and invariance. Both arise from the talker's overlapping productions of phonetic segments. Because segments are coarticulated, there are no boundaries between segments, and acoustic intervals dominated by properties of individual segments are context sensitive. Based on other perceptual research, an approach to the segmentation and invariance problems is proposed. I propose that listeners detect segment boundaries wherever acoustic information for individual segments begins or ends, and that thereby they preserve information about articulatory overlap; they do not impose boundaries between segments. That listeners respect articulatory overlap and use information for a phonetic segment within the acoustic domain of a predecessor is shown by recent research. This approach has implications for the invariance problem. If listeners use anticipatory information for a segment A within the acoustic domain of B as A information, perhaps the information does not, then, contribute to the percept of B. If listeners indeed “factor” A information from B, B will sound invariant over different As although its acoustic domain is context sensitive. I will present research in support of this hypothesis. [Work supported by NSF Grant BNS-8111470.]