This article discusses the implications of phonological representation for the study of L2 speech acquisition. It is argued, on the basis of empirical findings from diverse phenomena in L2 phonology, that refined representations in which 'segments' have internal prosodic structure offer a more insightful view of cross-linguistic phonetic interaction than traditional phonological models. These refinements may be implemented in the Onset Prominence representational environment, in which diverse structural parses affect sub-segmental phonetic properties, transitions between segments, and the formation of prosodic boundaries.