This article provides a parameter-free measurement of utility in intertemporal choice and presents new and more robust evidence on the discounting of money outcomes. Intertemporal utility was concave for gains and convex for losses, consistent with a hypothesis put forward by Loewenstein and Prelec (1992). Discount rates declined over time but less so than previously observed under the assumption of linear utility. For approximately 40% of our subjects constant discounting provided the best fit. The remaining 60% were most consistent with Harvey's (1986) power discounting. Our data provide little support for the popular quasi-hyperbolic model, which is widely used in economics today. We observed an asymmetry in the discounting of gains and losses that, unlike earlier findings, cannot be explained by a framing effect.