Using time-frequency-scale elements obtained from wavelet packets as a basis, we describe a broad framework of analysis which can be used to reveal the essential dynamics, identified as coherent structures, of precipitation. We show that the matching pursuits algorithm with nearly symmetric orthogonal wavelets provides an optimal representation of the inner structure of rainfall time series and can describe features that range from scales of isolated singularity to synoptically forced large-scale features. We describe the analysis of time series of several storms and show that there exist distinct scales of variation identifiable with rain cell and synoptic-scale activity, which is in contradistinction to the scale invariance hypothesis. ¿ American Geophysical Union 1996