Analyses of spatial patterns of distribution of populations and assemblages along envirorimental gradients are common in marine ecology. How these patterns vary at different spatial scales has seldom been examined, despite the fact that patterns in nature are intrinsically scale-dependent. This study quantified variability in subtidal assemblages at a hierarchy of spatial scales along a depth gradient, using several univariate and multivariate techniques. Despite variation in the sizes of depth effects in time and space, there were large, significant and generally characterisable differences in the structure of assemblages at different depths. The sizes of multivariate and univariate components of variation at different spatial scales were compared at each of 3 different depths (5, 15 and 25 m), using a bias-corrected bootstrapping approach. The sizes of variance components at different spatial scales varied with depth and choice of transformation. In all cases, the largest component of variation was at the smallest scale (tens of centimeters). A pattern of decreasing residual variance with depth was seen for untransformed data, while a pattern of increasing residual variance with depth was seen for presence/absence data. In contrast, variation among locations (separated by > 1 km) and among sites (separated by hundreds of metres) was largest at intermediate depths (similar to 15 m), regardless of the transformation used. The multivariate procedures used here offer several advantages over previously used techniques, providing suitable quantitative methods for analysing, at multiple scales, the patchy and complex nature of rocky subtidal assemblages.