Periphyton is so highly variable that the community resists modelling. To explore the unpredictability of this important resource, we studied: 1) alternative variables to the traditional trophic ones, 2) the spatial scale of periphyton variability, and 3) different levels of description of the community. For such an exercise, we partitioned the variability of epiphytic communities growing in Lake St. Francois (Quebec) into various fractions explained by environmental variables, spatial structure, and covariation between environment and space. We described the epiphytic communities according to their fine (61 taxa) or coarse (6 classes) taxonomy, their size structure, and their total biomass to identify which of these state variables could be most successfully modelled. Our analysis suggests that periphyton modelling could be improved by: 1) the measurement of physical and biotic factors as well as nutrients, 2) the consideration of microscale variation, and 3) the division of periphyton into functional gr...