OBJECTIVE: The reliability of electrocardiogram interpretation to diagnose myocardial ischemia in critically ill patients is unclear. In adults with septic shock, we assessed intra- and inter-rater agreement of electrocardiogram interpretation, and the effect of knowledge of troponin values on these interpretations. DESIGN: Prospective substudy of a randomized trial of vasopressin vs. norepinephrine in septic shock. SETTING: Nine Canadian intensive care units. PATIENTS: Adults with septic shock requiring at least 5 µg/min of norepinephrine for 6 hrs. INTERVENTIONS: Twelve-lead electrocardiograms were recorded before study drug, and 6 hrs, 2 days, and 4 days after study drug initiation. MEASUREMENTS: Two physician readers, blinded to patient data and group, independently interpreted electrocardiograms on three occasions (first two readings were blinded to patient data; third reading was unblinded to troponin). To calibrate and refine definitions, both readers initially reviewed 25 trial electrocardiograms representing normal to abnormal. Cohen's Kappa and the φ statistic were used to analyze intra- and inter-rater agreement. RESULTS: One hundred twenty-one patients (62.2 ± 16.5 yrs, Acute Physiology and Chronic Health Evaluation II 28.6 ± 7.7) had 373 electrocardiograms. Blinded to troponin, readers 1 and 2 interpreted 46.4% and 30.0% of electrocardiograms as normal, and 15.3% and 12.3% as ischemic, respectively. Intrarater agreement was moderate for overall ischemia (κ 0.54 and 0.58), moderate/good for "normal" (κ 0.69 and 0.55), fair to good for specific signs of ischemia (ST elevation, T inversion, and Q waves, reader 1 κ 0.40 to 0.69; reader 2 κ 0.56 to 0.70); and good/very good for atrial arrhythmias (κ 0.84 and 0.79) and bundle branch block (κ 0.88 and 0.79). Inter-rater agreement was fair for ischemia (κ 0.29), moderate for ST elevation (κ 0.48), T inversion (κ 0.52), and Q waves (κ 0.44), good for bundle branch block (κ 0.78), and very good for atrial arrhythmias (κ 0.83). Inter-rater agreement for ischemia improved from fair to moderate (κ 0.52, p = .028) when unblinded to troponin. CONCLUSIONS: In patients with septic shock, inter-rater agreement of electrocardiogram interpretation for myocardial ischemia was fair, and improved with troponin knowledge.