Assessment of climate change policies requires aggregation of costs and benefits over time and across generations, a process ordinarily done through discounting. Choosing the correct discount rate has proved to be controversial and highly consequential. To clarify past analysis and guide future work, we decompose discounting along two dimensions. First, we distinguish discounting by individuals, an empirical matter that determines their behavior in models, and discounting by an outside evaluator, an ethical matter involving the choice of a social welfare function. Second, for each type of discounting, we distinguish it due to pure time preference from that attributable to curvature of the pertinent function: utility functions (of consumption) for individuals and the social welfare function (of utilities) for the evaluator. We apply our analysis to leading integrated assessment models used to evaluate climate policies. We find that past work often confounds different sources of discounting, and we offer suggestions for avoiding these difficulties. Finally, we relate the standard intergenerational framework that combines considerations of efficiency and distribution to more familiar modes of analysis that assess most policies in terms of efficiency, leaving distributive concerns to the tax and transfer system.