Electrocardiograms taken on 451 young soldiers during a nutrition survey were read independently by 5 different observers. Seventeen records were considered "abnormal" by one or more observers; there was good agreement on 6 of the most marked abnormalities, but lesser deviations from normality among the 11 remaining tracings led to rather wide differences in opinion. The same 17 tracings were reinterpreted by each observer without knowledge of his earlier diagnosis; there were between 2 and 9 reclassifications, according to observer. Thirty-four tracings designated as "borderline" by one or more readers presented further variations in interpretation. All observers agreed, however, on the normality of the remaining 400 records. Observer variation would be reduced if reports on electrocardiograms were made primarily in terms of objective description rather than diagnostic interpretation.