A general descriptive framework of speech timing is proposed using a hierarchical linear (elastic) representation of speech units, with a loose linkage between different dimensions (articulators). Inherent temporal properties of phonetic units are represented by their spring constants, and prosodic properties of utterance units are represented by additional springs associated with relevant substructures. For describing a process of speech utterance, an open‐ended structure can be defined by terminal forces in place of position specifications, continuously building up the utterance structure as the speaker produces (or preprograms) the speech signal. Some examples of data from microbeam data will be discussed from this point of view, using “iceberg” patterns as timing marks [O. Fujimura, “A Linear Model of Speech Timing,” inIlse Lehiste Festschrift, edited by L. Shockey and R. Channon (Foris Publications, to be published)].