To develop a decision rule to aid in the diagnosis of myocardial infarction, we evaluated clinical and ECG data on 540 adults treated in an urban hospital emergency room for acute chest pain. Of 62 (11.5%) patients who had acute infarctions, 54 were admitted to intensive care (sensitivity 87%); 103 of 478 patients without infarctions were also admitted to intensive care (specificity 78%). Thirty-four percent of all patients admitted had infarctions. Multivariate analysis identified only four clinical variables which carried independent information predicting infarction: two from the ECG and two from the clinical history. A predictive model based on these four variables had significantly greater specificity (86% vs. 78%, p = .003) and accuracy of overall patient classification (88% vs. 79%, p = .013) but somewhat lower sensitivity (81% vs. 87%, p = .46) than physician judgments. However, a decision rule which would have admitted to intensive care those patients with a high probability of infarction who were not admitted by the emergency room physicians, would have increased the sensitivity for detecting infarction to 95% with no appreciable decrease in specificity or yield of infarctions among patients admitted to intensive care.