Despite the dynamic nature of spatial pattern, the temporal variation of spatial structure of marine benthic assemblages is rarely assessed using several temporal scales. We quantified the variability of density and biomass of main benthic species in the intertidal soft-bottom flats at two bights in Chupa Inlet (Kandalaksha Bay, the White Sea). The data cover the 21-year period (1987–2008) of a long-term monitoring survey (1987–present) using a hierarchical sampling design with two temporal (year, season within a year) and three spatial scales (bights—7 km, stations within a bight—10–100 m, and replicate samples—10 s cm apart). We used nested ANOVA to test significance and variance components to compare the relative contribution of different scales of variability of density and biomass of 18 most occurring macrobenthic species. Some species demonstrated high large-scale variability, however, the majority showed high small-scale variability and residual variance. The interactive variability was at least as important as the temporal effects, indicating that the spatial pattern changes through time. The assemblages were more variable at small scales and more stable at larger scales. Potential implications for sampling design are discussed.