The current decline in coronary artery disease mortality (CAD) may be a result of a declining population risk or of a declining case-fatality rate. Information on incidence trends for myocardial infarction (MI) could be used to distinguish between these 2 possibilities. Hospital discharge codes for MI (ICDCM-410) could be used as a convenient proxy for incidence trends, provided that coding of hospital discharges is sufficiently accurate. To evaluate the accuracy of medical records coding of patients signed out with an acute MI code (ICDCM-410), we compared them to an independent cardiology surveillance study of all patients with acute MI admitted to a large county teaching hospital. Over a 12-month period, 110 patients were coded as ICDCM-410 by medical records, but only 67 of these were detected by cardiology surveillance. The charts of the 43 patients not detected by surveillance were reviewed. In none of the 43 was evidence of acute MI found. In 28 of the 43, the discharge summaries listed rule out MI or status post-MI readmitted for further diagnostic workup, but were miscoded as ICDCM-410. Twelve of the 43 patients had cardiac arrests but were coded as ICDCM-410, even though there was no evidence of MI. Therefore, erroneous coding of patients as acute MI (ICDCM-410) may conceal a true downward trend in the incidence of CAD.