Threat cues are capable of quickly capturing and maintaining visual attention. The extent to which this attentional bias is indicative of threat characteristics or linked to the prediction of reliable outcomes remains uncertain. Moreover, the process of attentional exploration is shaped by learning, which tends to be selective towards more accurate predictors of an event. In order to explore the interaction of attentional and learning processes during fear acquisition and reversal, we created a novel multiple-cue paradigm. Initially, participants were sequentially presented with single visual cues (threat, safety, and ambiguous), each differing in their predictiveness for an aversive unconditioned electrotactile stimulus. In randomly interspersed multiple-cue displays (ambiguous cues together with either threat or safety cues), we tested whether attentional exploration is biased towards the predictive components of the display or influenced predominantly by the threat value.To investigate the readjustment of expectations, cue-outcome associations were switched between visual cues and aversive outcomes in the second half of the experiment (reversal learning). Shock expectancy ratings and autonomic responses (heart rate, skin conductance) indicated that participants flexibly learned the associations of cues to unconditioned stimuli, but they exhibited a threat-related bias reflecting a better-safe-than-sorry strategy. On the other hand, eye movement data (latency of first fixation, proportional dwell time) revealed that attention is predominantly driven by uncertainty in cues. Reinforcement learning models will be used to map these results. Our research provides insights into how attentional and learning processes interact during fear acquisition and reversal, with implications for understanding threat-related biases in anxiety.