Summary1. Plant functional types (PFTs) are groups of species sharing traits that govern their mechanisms of response to environmental perturbations such as recurring fires, inundation, grazing, biological invasions and global climate change. The key components of a PFT approach are an underlying model of vegetation dynamics for a given system and a classification of functional types based on traits deduced from key processes in the model. 2. Prediction and generalization underpin the potential utility of the PFT approach for understanding ecosystem behaviour. For PFTs to be useful in ecosystem management, they (in concert with their underlying model) must reliably predict vegetation change under given environmental scenarios and they must produce robust generalizations across the species that are classified and the range of environments in which they occur. 3. The efficacy of plant functional types has been explored using various approaches in a wide range of ecosystems. However, very few studies have tested the accuracy and generality of PFT predictions against vegetation changes observed empirically over medium to long time scales. 4. We applied this approach to examine the predictive accuracy and generality of a PFT classification and an associated model of vegetation dynamics for a fire-prone, species-rich wet heathland in south-eastern Australia. We assigned each species to one of six PFTs derived using a deductive approach based on the vital attributes scheme. We measured their initial abundance at a set of sample sites distributed across local environmental gradients. We used the PFT traits and processes in the underlying model to predict qualitative changes in abundance in response to a fire regime scenario observed at the sample sites during a subsequent period of 21 years. We then re-surveyed the sample sites to compare predictions with observed changes in abundance. 5. The PFTs and their underlying model produced an accurate prediction of average vegetation responses over the 21-year period. The majority of species within each PFT exhibited the predicted response and few species had strongly opposing responses in different environments. However, not all species within a PFT underwent the predicted direction of change, and responses of individual species were not uniform across the environmental gradients. 6. Synthesis . We conclude that plant functional types based on vital attributes are very useful tools for prediction and generalization in ecosystem management, although interpretations need to be tempered by the fact that PFTs may not accurately predict responses of all species across all environments.