The authors describe the use of neural networks to recognize stenosis in patients with arteriographically proven absence or presence of coronary artery disease by analyzing of thallium-201 scintigrams. There are four possible classification categories: normal, stenosis of the nondominant circumflex coronary artery, stenosis of the left anterior descending, or stenosis of the right coronary artery, including the posterior descending. Two different neural network algorithms are applied to the scintigraphic data: the backpropagation method and the Kohonen algorithm. The backpropagation network consists of three layers of neurodes and utilizes the Delta rule as the learning rule. The Kohonen method involves a single layer of neurodes which uses an unsupervised learning rule to cluster the different classes of data. Both networks are trained using 70% of the scintigraphic data, with the remainder of the data used for testing. The results obtained by using backpropagation are fair with an accuracy of 66.6%. The results from the Kohonen algorithm are better, especially in terms of specificities for both abnormal and normal patients.<<ETX>>