The study of food webs has been a central theme within ecology for decades, and their structure and dynamics have been used to assess a range of key properties of communities (e.g. complexity–stability relationships) and ecosystems (e.g. fluxes of energy and nutrients). However, many food web parameters are sensitive to sampling effort, which is rarely considered, and further, most studies have used either species- or size-averaged data for both nodes and links, rather than individual-based data, which is the level of organisation at which trophic interactions occur. This practice of aggregating data hides a considerable amount of biologically meaningful variation and could, together with potential sampling effects, create methodological artefacts. New individual-based approaches could improve our understanding of, and ability to predict, food web structure and dynamics, particularly if they are derived from simple metabolic and foraging constraints. We explored the effect of species-averaging in four highly-resolved individual-based aquatic food webs (Broadstone Stream, the Afon Hirnant, Tadnoll Brook and the Celtic Sea) and found that it obscured structural regularities resulting from intraspecific size variation. The individual-based approach provided clearer insights into seasonal and ontogenetic shifts, highlighting the importance of the temporal component of size-structuring in ecological networks. An extension of the Allometric Diet Breadth Model predicted the structure of the empirical food webs almost twice as accurately as the equivalent species-based webs, with the best-fitting model predicting 83% of the links correctly in the Broadstone Stream size-based web, and the few mismatches between the model and data were explained largely by sampling effects. Our results highlight the need for theoretical explanations to correspond closely with methods of data collection and aggregation, which is the exception rather than the rule at present. We suggest how this situation can be improved by including individual-level data and more explicit information on sampling effort when constructing food webs in future studies.