To evaluate the ability of emergency room physicians to estimate the probability of myocardial infarction in patients with acute chest pain, the authors gathered historical, physical, and electrocardiographic information from 492 patients at the time of their presentation. The physicians admitted 30% of them to intensive care: 53 of the 61 patients with infarctions (sensitivity = 87%) and 96 of the 431 without infarctions (specificity = 78%). Overall, 36% of those admitted had infarctions. The physicians' numeric estimate of the probability of infarction was a good univariate discriminator of infarction, as demonstrated by Receiver Operator Characteristics analysis, and, as indicated by their actual operating point, they seemed to maximize the accuracy of patient classification rather than sensitivity or specificity. Logistic regression analysis identified the physicians' probability estimate as the strongest multivariate predictor of infarction, considering all other clinical information available.