This study reports three experiments which demonstrate path dependency in risky intertemporal choice. Consider a lottery to be resolved and paid in a future time period. One can obtain the present value of this lottery in three different ways: (1) eliciting directly the present certainty equivalent (CE) of the future lottery (direct path); (2) eliciting the future CE and then discounting this amount to the present (risk-time path); and (3) eliciting the present value of the risky prospect and then determining the CE of this current lottery (time-risk path). Standard rational choice models such as the discounted expected utility model, assume a multiplicative model, where all three methods mentioned above would yield the same value. We conducted three studies to examine if this is the case: Experiments 1 and 2 were based on a set of matching-task questions and Experiment 3 used a process-tracing design to analyze the natural sequence of decision making by the subjects. These three studies show that the evaluation of future gambles is path-dependent. The present values elicited under the time-risk and direct paths are, on average, higher than those reported under the risktime path. In addition, we found evidence for a two-stage evaluation of risky future prospects: When evaluating a future gamble, individuals first assess the present value of the gamble (time discounting) and then they determine a certainty equivalent (probability discounting).