AbstractCell size is an important determining factor for predicting the physiological and ecological properties of phytoplankton. Size dependencies in eco-physiological properties are in general reported as log–log linear scaling relationships. Considerable uncertainty in these allometries, hence, limits our ability to link them to observed changes in phytoplankton community structures. In this study, we develop a size-based multi-species phytoplankton model and assess the sensitivity of the predicted community size structure to variations in allometric coefficients. The model describes the nutrient–phytoplankton–detritus dynamics within the upper mixed layer for a matrix of habitats, which are characterized by the deep layer nutrient concentration and mixing frequency. Predicted diatom community mean cell size becomes maximal at intermediate mixing frequencies, which confirms the importance of storage capacity (relative to the subsistence demand) at intermittent nutrient supplies. Smaller subsistence demand of large diatoms makes a critical factor in shaping the community size structure, while in environments with either short or long nutrient replenishment periods, maximum growth rate gains similar or more importance. Notably, in these environments, the diatom community converges towards unrealistically small species when we assumed a uniform (log–log linear) allometry in maximum growth rate. Independent theoretical and empirical arguments motivated the usage of non-uniform growth scaling, with which the minimal diatom cell size actually observed in nature is realized in long-term simulations. Using allometries parameterized for the entire phytoplankton community, the subsistence advantage of the larger species becomes insignificant, leading to a ubiquitous dominance of smaller species, even when assuming non-uniform scaling in maximum growth rate. This finding corresponds with the observed dominance of pico-phytoplankton in many parts of the ocean. All combinations of physiological allometries for diatom or mixed communities, however, underestimate both the mean cell size of the community and also size diversity. This may indicate a significant role of other ecological selection mechanisms, such as arising from size-selective grazing. The approach outlined in this paper helps to better assess the limits and the potential of size based phytoplankton models as an increasingly important tool in plankton research.