The goal of this study was to determine how the fear relevancy of outcomes during probabilistic classification learning affects behavior and strategy use. Novel variants of the "weather prediction" task were created, in which cue cards predicted either looming fearful or neutral outcomes in a between-groups design. Strategy use was examined by goodness-of-fit estimates of response patterns across trial blocks to mathematical models of simple, complex, and nonidentifiable strategies. Participants in the emotional condition who were fearful of the outcomes had greater skin conductance responses compared with controls and performed worse, used suboptimal strategies, and had less insight into the predictive cue features during initial learning. In contrast, nonfearful participants in the emotional condition used more optimal strategies than the other groups by the end of the two training days. Results have implications for understanding how individual differences in fear relevancy alter the impact of emotion on feedback-based learning.Learning from emotional experiences is an important survival skill across species. Environmental contingencies predicting negative or positive outcomes provide key information useful for assessing the motivational value of selected actions and to assist decision-making processes in guiding future behavior. One form of contingency learning involves the gradual acquisition of cueoutcome associations guided by feedback (procedural or habit learning) (Mishkin et al. 1984). While much research has examined the cognitive and neural mechanisms underlying this form of learning, the influence of emotion has not been systematically addressed as it has for other domains of memory (LaBar and Cabeza 2006). Here we examine how individual differences in fear relevancy modulate behavioral performance and strategy use on a probabilistic classification learning (PCL) task that involves trial-and-error learning of associations between cues and outcomes that vary in emotional salience.In the standard version of a PCL task (the "weather prediction" task) (Knowlton et al. 1994), participants predict the weather in a foreign city (rain or sunshine) based on the presence of a combination of four cue cards. Across training, participants learn to probability match the appearance of the cue cards by choosing the outcome with the same probability that they are reinforced. Given that individuals tend to have little insight into their performance, there has been interest in characterizing the underlying response patterns, or strategies used to solve the task. To accomplish this objective, each participant's data is mathematically modeled across trial blocks to determine the goodness-of-fit to an "ideal" responder following particular response patterns. Least-mean-square estimates indicate that participants use at least three classes of strategies varying in optimality: (1) no identifiable strategy, (2) simple strategies involving the use of one cue to make predictions, and (3) complex strategies involving the use o...