Panic disorder is characterized by the paroxysmal occurrence and fear of bodily symptoms. In recent years it has been proposed that patients “learn” to fear cardiorespiratory sensations through interoceptive conditioning. This study sought to model the initial stage of this process in healthy volunteers (N=44) using mild cardiac sensations. An additional aim was to explore whether anxiety sensitivity – a known risk factor for panic disorder – modulates such interoceptive learning. Infusions of pentagastrin and saline were used to manipulate the presence versus absence of cardiac sensations, respectively, and served as conditioned stimuli in a differential interoceptive conditioning paradigm. Inhalation of 35% CO2-enriched air served as the panicogenic, unconditioned stimulus (UCS). In half of the participants (“prepared” condition), cardiac sensations caused by pentagastrin were followed by inhalation of CO2-enriched air (penta CS+), whereas the absence of such sensations (saline) was followed by room air (saline CS−). The reversed combination (“unprepared” condition) was used in the other half of the participants. Conditioning effects showed up for self-reported UCS-expectancy, but not for skin conductance and anxiety ratings. Only participants from the prepared group learned to expect the UCS, and differential learning was impaired with higher scores on anxiety sensitivity. Expectancy learning was more easily established towards the presence compared to the absence of cardiac sensations, whereas the reverse effect was observed for safety learning. Modeling impaired discriminatory learning and the moderating effect of anxiety sensitivity provides new insight in the development of panic disorder.