Psychoevolutionary theories postulate a threat-selective, evolutionary-based fear learning mechanism responsible for faster and more persistent learning towards stimuli that have threatened human survival. However, research findings based on appraisal theories reveal that these learning biases are not restricted towards this type of stimuli, but also extend towards positive stimuli that are evaluated as relevant to an individual’s concerns in aversive Pavlovian conditioning (Stussi et al., 2018, 2021). Here, the present study aims to replicate these results by using the measure of pupil dilation as an index of the conditioned response (CR). To do so, a differential aversive Pavlovian paradigm (N = 56) was used to test the hypotheses that positive and negative relevant stimuli are more rapidly and persistently conditioned to threat, compared to neutral stimuli. Angry male faces, baby faces and neutral male faces were used as conditioned stimuli and an electric stimulation as the unconditioned stimulus. To measure the participants’ CR, pupil dilation was used. In line with the first hypothesis, the CR towards angry faces and baby faces was more readily acquired compared to neutral faces. Conversely, we did not observe a more persistent CR towards affectively relevant versus neutral stimuli. Together, the present study provides mixed evidence for a general relevance detection mechanism underlying affective learning biases in humans.