To test the power of the 35% CO2 test as a challenge for panic disorder (PD), and to set an ideal threshold of discrimination between patients and controls, we analysed by receiver operating characteristic (ROC) analysis the responses of 91 out-patients with PD and 46 controls who inhaled a 35% CO2/65% O2 gas mixture. ROC analysis confirmed that the CO2 challenge discriminates well between PD patients and controls, with 86% probability to classify them correctly on the basis of subjective anxiety after the test. A relatively modest increment in subjective anxiety (i.e. an absolute increment of 20 units, or a relative increment of 26% of subjective anxiety) proved to be the ideal threshold to separate the two groups of our sample.