Background: The purpose of this study was to measure the degree to which patients and their treating physicians correctly guessed whether patients were on an active treatment (paroxetine or clomipramine) or pill-placebo and whether correctness of these guesses was related to treatment outcome. Methods: Ninety-five panic disorder patients, randomized to receive double- blind treatment with paroxetine, clomipramine or placebo for twelve weeks, were asked half-way through this period to classify treatment as active or placebo. Medical doctors were asked the same. Results: Both patients and physicians guessed correctly to a degree much greater than would be expected by chance whether the patient was on an active treatment. Neither patients nor physicians were good at estimating correctly whether a patient was on a placebo. There was a trend approaching significance for patients on a placebo, whose physicians believed that they were on active treatment, to have a higher rating of symptom improvement than those patients who were correctly guessed to be on placebo. Conclusion: The 'double-blindness' procedure did not mask the giving of antidepressive medication in panic disorder. There is some evidence that physicians who incorrectly classify patients on a placebo as receiving active treatment relate this to better treatment outcome.