To evaluate how well a blind is maintained in a double-blind study. Clinicians (n = 66), parents (n = 62), and depressed child/adolescent subjects (n = 62) predicted whether the patient had been on either placebo or active medication at the end of an eight-week double-blind placebo versus fluoxetine trial. Clinician, patient and parents' guesses as to which treatment they had received were at a chance level based on an overall analysis. However, when clinical response and condition assignment were controlled, all were correctly predicting placebo treatment but not medication treatment. The finding that subjects, parents and clinicians predict at a chance level is important for double-blind study design integrity. However, clinicians, parents and subjects were accurately predicting placebo treatment when clinical response and the assigned condition were taken into account but not medication. Since they do not know condition however, all remain essentially blinded, and this is an important finding for design and analysis integrity for double-blind studies.