Prediction is an important aspect of scientific practice, because it helps us to confirm theories and effectively intervene on the systems we are investigating. In Ecology, prediction is a controversial topic: even though the number of papers focusing on prediction is constantly increasing, many ecologists believe that the quality of ecological predictions is unacceptably low, in the sense that they are not sufficiently accurate sufficiently often. Moreover, ecologists disagree on how predictions can be improved. On one side are the ‘theory-driven’ ecologists, those who believe that ecology lacks a sufficiently strong theoretical framework. For them, more general theories will yield more accurate predictions. On the other are the ‘applied’ ecologists, whose research is focused on effective interventions on ecological systems. For them, deeper knowledge of the system in question is more important than background theory. The theory-driven approach has sound philosophical commitments, but is not best suited to ecological systems, as they are not merely complex but also causally heterogeneous. Thus, there is a tradeoff between generalisability and predictive power. I am more sympathetic to the opposing view, as it is better suited to the reality of ecological systems, though I believe that some of its proponents mischaracterise ‘applied’ predictions and their role in ecological research. I will propose an alternative approach for classifying ecological predictions that may further improve predictive ecology.