It is well-established that anxiety and/or depression are associated with a negative bias when interpreting ambiguous information. This study tested the novel hypothesis that the criterion one sets for judging a stimulus as threatening is a core aspect of this bias. A sample of 174 participants were divided into neutral (n = 87) and threatening (n = 87) training conditions. Participants performed a facial expression detection task, in which criterion was shifted in the liberal (threatening condition) or conservative (neutral condition) direction via differential reward contingencies. Training conditions were successful in inducing large shifts in criterion as intended. There was also a small change in sensitivity in the neutral condition, however, the manipulation is still considered successful given the substantive effect size for change in criterion compared to change in sensitivity. As predicted, conservative criterion-training resulted in significantly lower levels of negative affect post-training. No significant change was found for liberal criterion-training on negative affect. Positive affect also decreased across time regardless of condition. Overall, the reduction in negative affect following conservative criterion-training demonstrates that modifying criterion impacts affect and identifies criterion setting as a potential target in the treatment of mental health disorders with prominent negative affect.