Prediction of community dynamics remains challenging, but important for a range of conceptual and applied ecological contexts (e.g. species invasions, sustainable harvesting, restoration planning). Functional traits offer a potential solution via their often-strong linkages with environmental variables – perhaps community dynamics can be predicted by determining the species with appropriate traits for a given environment. But do the conceptual arguments hold up, and does the approach work in practice? I will present our ongoing work around the limits to predictability, comprising a synthesis of theory for community response to environmental change, contemporary demographic studies of whole plant communities, and paleoecological analysis of plant community dynamics over the Quaternary. Our work is showing that environmental change often should not yield simple community responses, but that there are trait-based approaches that may offer hope of predictability.