To determine whether data available to physicians in the emergency room can accurately identify which patients with acute chest pain are having myocardial infarctions, we analyzed 482 patients at one hospital. Using recursive partitioning analysis, we constructed a decision protocol in the format of a simple flow chart to identify infarction on the basis of nine clinical factors. In prospective testing on 468 other patients at a second hospital, the protocol performed as well as the physicians. Moreover, an integration of the protocol with the physicians' judgments resulted in a classification system that preserved sensitivity for detecting infarctions, significantly improved the specificity (from 67 per cent to 77 per cent, P less than 0.01) and positive predictive value (from 34 per cent to 42 per cent, P = 0.016) of admission to an intensive-care area. The protocol identified a subgroup of 107 patients among whom only 5 per cent had infarctions and for whom admission to non-intensive-care areas might be appropriate. This decision protocol warrants further wide-scale prospective testing but is not ready for routine clinical use.