Eighteen subjects meeting DSM-III-R criteria for panic disorder with agoraphobia were randomly allocated to either a safety or non-safety condition and underwent 15 min inhalations of 5.5% CO2 in air. In the safety condition, both a senior and junior experimenter were present, the junior experimenter was presented as a professional, and subjects were in personal contact with the senior experimenter throughout the procedure. In the non-safety condition, the senior experimenter left on a bogus emergency, the junior experimenter was introduced as a "student" and was dressed casually, and subjects were alone in the experimental room throughout the procedure. Despite these dramatic manipulations, there was no significant difference between groups on either the number of subjects experiencing a panic attack or on pre-inhalation safety scores. However, a median-split analysis on safety scores indicated that those subjects who perceived themselves to be less safe before the inhalation were more likely to experience a panic attack.