Noninvasive testing is often evaluated by the sensitivity and specificity in comparison with a more invasive, but more definitive "gold" standard. However, work-up or verification bias, which occurs when the results of a noninvasive test impact the decision to perform the gold standard invasive test, increases the "observed" sensitivity and decreases the "observed" specificity of the noninvasive test. Most large clinical studies utilizing a noninvasive technique to diagnose coronary artery disease have biases, particularly work-up bias. To obtain more accurate measurements of sensitivity and specificity, we determined the observed sensitivity and specificity of stress (exercise and dipyridamole) single photon emission computed tomographic (SPECT) thallium testing for the detection of coronary artery disease by angiography, and then applied previously published equations to correct for work-up bias. From a computerized data base, reports of 4354 stress SPECT thallium studies from January 1, 1986 through December 31, 1992 were reviewed. All patients with a known history of myocardial infarction or prior coronary angiography were excluded, leaving 2688 patients. From this total, 471 patients underwent coronary angiography within 90 days following stress SPECT thallium testing. Coronary artery disease was defined as a visually assessed stenosis of a coronary artery or a major branch > 50%. Of the 2688 stress SPECT thallium studies, 1265 were normal and 1423 were abnormal. For the 471 patients who underwent catheterization within 90 days following stress SPECT thallium testing. the "observed" sensitivity and specificity were 98 and 14%, respectively. After correction for work-up bias, the corrected sensitivity and specificity were 82 +/- 6% and 59 +/- 2%, respectively. Most studies utilizing noninvasive technologies for the detection of coronary artery disease include patients with known coronary artery disease and have work-up bias as well. By knowing the thallium results of patients with and without catheterization, we were able to correct for work-up bias. These data provide better estimate of the sensitivity and specificity of stress SPECT thallium testing.