How can children use prosodic information such as pausing and pitch resetting to infer syntactic structure? This work considers the possibility that they do so by first constructing a prosodic representation similar to ones suggested by recent linguistic theory (e.g., Hayes, 1989; Nespor & Vogel, 1986; Selkirk, 1981). The research examines 2-year-olds′ sensitivity to prosodic structure by comparing their weak syllable preservation patterns in comparably stressed multisyllabic words and sentences. Consistent with previous research, the results suggest a speech production model in which children apply to their multisyllabic words a series of strong-(weak) metrical templates (Allen & Hawkins, 1980; Demuth, 1992, in press; Gerken, 1990, 1991, in press b; Gerken, Landau, & Remez, 1990; Wijnen, Krikaar, & den Os, in press). The new finding added by the current research is that children organize their intended sentences into phonological phrases and apply metrical templates within these prosodic units. Implications of children′s sensitivity to prosodic structure for syntax acquisition are discussed.