Preliminaries 1Several recent investigations of the development of left-edge clusters in West Germanic languages have demonstrated that the relative sonority of adjacent consonants plays a key role in children's reduction patterns (e.g. Fikkert 1994, Gilbers & Den Ouden 1994, Chin 1996, Barlow 1997, Bernhardt & Stemberger 1998, Gierut 1999, Ohala 1999.These authors have argued that, for a number of children, at the stage in development when only one member of a left-edge cluster is produced, it is the least sonorous segment that survives, regardless of where this segment appears in the target string or the structural position that it occupies (head, dependent, or appendix). To briefly illustrate, while the more sonorous /S/ 2 is lost in favour of the stop in /S/+stop clusters, /S/ is retained in /S/+sonorant clusters; similarly, the least sonorous stop survives in both /S/+stop and stop+sonorant clusters, in spite of the fact that it occurs in different positions in the two strings. To account for reduction patterns such as these, a structural difference between /S/-initial and stop-initial clusters need not be assumed. This would seem to fare well in view of much of the recent constraint-based literature which de-emphasises the role of prosodic constituency in favour of phonetically-based explanations of phonological phenomena (see e.g. Hamilton 1996, Wright 1996, Kochetov 1999, Steriade 1999, Côté 2000.In this paper, we focus on a second set of reduction patterns for left-edge clusters, one which is not addressed in most of the sonority-based literature on cluster reduction in child language (L1): these patterns reveal a preference for structural heads to survive.