The present experiment examined the possibility to change the sensory and/or the affective perception of thermal stimuli by an emotional associative learning procedure known to operate without participants' awareness (evaluative conditioning). In a mixed design, an aversive conditioning procedure was compared between subjects to an appetitive conditioning procedure. Both groups were also compared within-subject to a control condition (neutral conditioning). The aversive conditioning was induced by associating non-painful and painful thermal stimuli - delivered on the right forearm - with unpleasant slides. The appetitive conditioning consisted in an association between thermal stimuli - also delivered on the right forearm - and pleasant slides. The control condition consisted in an association between thermal stimuli - delivered for all participants on the left forearm - and neutral slides. The effects of the conditioning procedures on the sensory and affective dimensions were evaluated with visual analogue scale (VAS)-intensity and VAS-unpleasantness. Startle reflex was used as a physiological index of emotional valence disposition. Results confirmed that no participants were aware of the conditioning procedure. After unpleasant slides (aversive conditioning), non-painful and painful thermal stimuli were judged more intense and more unpleasant than when preceded by neutral slides (control condition) or pleasant slides (appetitive conditioning). Despite a strong correlation between the intensity and the unpleasantness scales, effects were weaker for the affective scale and, became statistically non-significant when VAS-intensity was used as covariate. This experiment shows that it is possible to modify the perception of intensity of thermal stimuli by a non-conscious learning procedure based on the transfer of the valence of the unconditioned stimuli (pleasant or unpleasant slides) towards the conditioned stimuli (non-painful and painful thermal stimuli). These results plead for a conception of pain as a conscious output of complex informational processes all of which are not accessible to participants' awareness. Mechanisms by which affective input may influence sensory experience and clinical implications of the present study are discussed.