Good translatability of behavioral measures of affect (emotion) between human and nonhuman animals is core to comparative studies. The judgment bias (JB) task, which measures “optimistic” and “pessimistic” decision-making under ambiguity as indicators of positive and negative affective valence, has been used in both human and nonhuman animals. However, one key disparity between human and nonhuman studies is that the former typically use secondary reinforcers (e.g., money) whereas the latter typically use primary reinforcers (e.g., food). To address this deficiency and shed further light on JB as a measure of affect, we developed a novel version of a JB task for humans using primary reinforcers. Data on decision-making and reported affective state during the JB task were analyzed using computational modeling. Overall, participants grasped the task well, and as anticipated, their reported affective valence correlated with trial-by-trial variation in offered volume of juice. In addition, previous findings from monetary versions of the task were replicated: More positive prediction errors were associated with more positive affective valence, a higher lapse rate was associated with lower affective arousal, and affective arousal decreased as a function of number of trials completed. There was no evidence that more positive valence was associated with greater “optimism,” but instead, there was evidence that affective valence influenced the participants' decision stochasticity, whereas affective arousal tended to influence their propensity for errors. This novel version of the JB task provides a useful tool for investigation of the links between primary reward and punisher experience, affect, and decision-making, especially from a comparative perspective.