We present an adaptation of existing variance partitioning methods todecompose species-environment relationships in hierarchically-structured,multi-scaled data sets. The approach translates a hierarchical, multi-scaleconceptual model into a statistical decomposition of variance. It uses a seriesof partial canonical ordinations to divide the explained variance inspecies-environment relationships into its independent and confoundedcomponents, facilitating tests of the relative importance of factors atdifferent organizational levels in driving system behavior. We discuss themethod in the context of an empirical example based on forest bird communityresponses to multiple habitat scales in the Oregon Coast Range, USA. Theexamplepresents a two-tiered decomposition of the variation in the bird community thatis explainable by a series of habitat factors nested within three spatialscales(plot, patch, and landscape). This method is particularly suited for theproblems of hierarchically structured landscape data. The explicit multi-scaleapproach is a major step forward from conducting separate analyses at differentscale levels, as it allows comprehensive analysis of the interaction of factorsacross scales and facilitates ecological interpretation in theoretical terms.