Informed consent has been indirectly studied only in settings that do not replicate the actual consent process. We designed a sham study and randomly allocated adult ambulatory patients to receive one of two consent forms: Consent A (n = 52) described a randomized trial of usual treatment vs a new medication that "may work twice as fast as the usual treatment"; or Consent B (n = 48) that described a randomized trial of a new medication that "may work half as fast as the usual treatment". Patients randomized to Consent A were more likely to consent than those randomized to Consent B (consent rate A = 67%, consent rate B = 42%, p less than 0.01). Among patients who cited quantitative information, the difference in consent rate was even more marked (95% vs 36%, p less than 0.001); patients who did not cite quantitative information had equivalent consent rates. Patients who perceived minimal or severe symptoms had lower consent rates than those with mid-range symptom scores (chi 2(2) = 8.35, p = 0.015). Patients who recognize quantitative information will use it to make informed consent decisions.