This paper deals with risk sensitivity in amount of food, and is concerned with modelling the risk-sensitive behaviour exhibited by an animal which is attempting to survive a period such as winter. I argue that, in maximising survival probability, risk-averse behaviour is much more important than risk-prone behaviour. I also argue that animals in the laboratory will continue to use rules which are adapted to food sources which change over time. A model of environmental change is investigated. This model predicts that a consequence of change is that less riskprone behaviour than that predicted by standard models is to be expected in laboratory studies. The need to learn about a stochastic food source in a changing environment is predicted to further reduce the incidence of risk-prone behaviour. It is difficult to learn about highly variable food sources. This is shown to lead to risk-aversion even when the optimisation criterion is maximisation of mean long-term rate of food gain. Implications of the general modelling philosophy are discussed.