The "subjective value" computation assumption holds that intertemporal choice is performed by computing subjective value by eliciting information searching within alternatives. Although widely accepted, this assumption has not been well tested. In the current eye tracking study, we developed a new paradigm using forward inference to examine this hypothesis. Participants were instructed to finish two types of intertemporal choice tasks. The free task resembled a standard task, in which participants were free to follow their own preference. The instructed task, on the other hand, provided a baseline whereby participants were given explicit instructions to calculate the subjective value of options according to a classic discounting model. The results of behavioral, fixation and scanpath demonstrated that participants performed the instructed task in a manner that was consistent with a discounting computation process. However, by comparing performances across the two tasks, we found that participants made more attribute-based saccades when they made instructed choices, compared to when they made free choices. These findings provide no support for the "subjective value" assumption in intertemporal choice. That is, that the computation of subjective value is associated with alternative-based information searching, further demonstrating the effectiveness of the forward inference paradigm in testing intertemporal choices models. This study also highlights the need for careful reexamination of the assumed relationship between eye movement patterns and corresponding underlying cognitive processes in intertemporal choice.