This paper argues that observations of non-stationary choice behavior need not necessarily imply specific properties of the individual's discount function. As we show, the observed "anomalies" in intertemporal choice can alternatively be explained by an individual's perception of the risk that is involved whenever an outcome is to be received in the future. This risk may concern the size of the actual outcome or the endowment consumption stream to which the outcome is added. Both types of uncertainty naturally appear in the context of intertemporal choice and both are difficult to control in experiments. We show how relative degrees of changes in risk over time can predict choices.