In the era of health care reform, third party health payors are using decision support systems to justify the use of costly invasive procedures. In the field of cardiology, a technology that could predict significant coronary artery disease (CAD) from non-invasive data would be helpful. Artificial neural networks (ANN) have the potential to recognize subtle patterns in data and apply its training to a new set of data. We trained a personal computer-based ANN, using back propagation of errors through 23 hidden neurons, incorporating logistic transfer function, in a single layer, to identify patients with significant CAD based on non-invasive data. We trained the ANN by using non-invasive data from 276 consecutive patients who underwent cardiac catheterization. Input data included age, sex, history, physical exam, cardiac risk factors, exercise stress test data (with or without nuclear imaging). The ANN was trained to recognize significant CAD, defined as g 50% obstruction of the left main or other major epicardial vessels, The ANN was then applied to patients not used in training. Significant CAD was identified with a positive predictive accuracy of 80%, and a negative predictive accuracy of 92%. In conclusion, ANN can exclude patients without significant CAD with a high degree (92%) of confidence. This technique has the potential to limit the use of costly invasive procedures.