OBJECTIVE: Previous studies have suggested that manipulations in patient instructions before inhalation of carbon dioxide (CO2) may blunt its anxiogenic effects. The authors examined the effects of a range of instructional types on panic rates and physiological responses to CO2 inhalation. METHOD: Thirty-seven patients with panic disorder and 16 normal subjects were exposed to CO2 inhalation under one of three instructional conditions: 1) reassurance (reiterated safety assurances), 2) control dial (patients were instructed on using an operative dial to reduce the concentration of CO2 in the breathing canopy), and 3) basic instructions (standard information). Perceived breathlessness was measured with the Borg Scale of Respiratory Exertion. RESULTS: Manipulation of instructional sets did not affect panic rates among patients. The reassurance condition, however, produced a significant reduction in perceived breathlessness. CONCLUSIONS: CO2-induced panic is a robust biological effect that occurs independent of cognitive set changes, such as illusion of control or reassurance of safety. Conclusions drawn from this study are bolstered by the fact that it is methodologically rigorous: it had a large number of subjects, used random assignment to one of three instructions, included a full array of both subjective and biological measures, and included independent ratings.