Ecology is developing into a quantitative science, following the success of quantitative models in physics. Physics, especially classical mechanics, represents the paradigm of a ‘hard science’. It elaborates general theories and quantitative models. These models describe mechanical processes precisely and produce reliable predictions in their field. Nevertheless, physics and ecology study different types of subjects. The authors believe that the kinds o f uncertainty ecology deals with are inherent to the biotic processes and therefore partially irreducible: we cannot that scientific progress will reduce uncertainty to any intended level. This inherent and irreducible uncertainty will prevent ecology from achieving the generality and predictability o f mechanics. Therefore, strategies and techniques to deal with ecological uncertainty are needed and desirable. For administration and management, a general purpose of science is to reduce uncertainty in decision making. We need to know the potential sources and features o f different types of uncertainty, and the limitations o f current ecology. Mathematical models are major quantitative tools to produce ecological inferences (Jorgensen 1988). Many different kinds of quantitative models exist, including analytical models, simulation models, and statistical models (Loehle 1983, Haefner 1996, Hilbom and Mangel 1997). Each of them have specific merits and drawbacks in handling uncertainties. Ecologists often favour one of them over another. Uncertainty trade-offs exist in ecological modelling and have rarely been addressed. Environmental decision makers and mangers need practical strategies to handle uncertainty in ecological knowledge and ecological modelling and to be prepared for unexpected events. This paper will show the sources and features of uncertainty in ecology and particularly in ecological modelling and suggest strategies to handle ecological uncertainty. The authors suggest, we should use plural, multiple and linked approaches in integrative frameworks to address uncertainty problems.