Understanding how outcome affective value influences Pavlovian learning is critical for elucidating the mechanisms underlying human emotional learning. In this study, participants completed a Pavlovian conditioning task involving outcomes varying in their level of aversiveness: a tactile (nonaversive) or painful (aversive) shock, and no-shock. Skin conductance responses (SCRs), explicit pleasantness ratings of each CS, and CS-outcome contingency ratings were recorded. Although both painful and tactile outcomes elicited unconditioned SCRs, with greater responses for the former, only the painful outcome triggered anticipatory conditioned SCRs and robust prediction error-related SCRs following its unexpected omission. Computational modeling, through a Rescorla-Wagner model, revealed that prediction errors estimated by the model significantly predicted SCRs following painful outcome omissions, confirming their role as prediction error signals. In addition, the model-derived outcome sensitivity parameter, reflecting the affective value assigned to the different outcomes by each participant, was higher for painful than tactile stimuli, effectively discriminating between them, whereas learning rates did not differ across outcomes. These findings refine our understanding of emotional learning by showing how the affective value of the outcome shapes the predictive mechanisms underlying conditioned responding, as revealed through converging autonomic, explicit, and computational modeling evidence.