The aim of this study was to compare the accuracy of pulse dye densitometry with that of bolus thermodilution cardiac output measurement in patients before and after elective coronary artery bypass grafting. Twenty-eight patients were studied. Agreement between mean thermodilution and pulse dye densitometry cardiac output values was assessed by Bland-Altman analysis. Preoperative median [range] cardiac output was 3.87 [2.37–6.0] l.min−1 by thermodilution, and 3.11 [1.7–5.45] l.min−1 by pulse dye densitometry using indocyanine green 5 mg. Pulse dye densitometry underestimated cardiac output (mean bias − 0.42 l.min−1); the limits of agreement were ± 1.91 l.min−1, and mean error was 50.3%, indicating low precision. Preoperative median [range] cardiac output was 3.85 [2.2–6.0] l.min−1 for bolus thermodilution cardiac output and 4.2 [2.0–7.2] l.min−1 for pulse dye densitometry using indocyanine green 20 mg. Mean bias was + 0.566 l.min−1, the limits of agreement were ± 2.51 l.min−1 and mean error was 60.9%. Postoperative cardiac output data were not analysed because pulse dye densitometry signals were low or absent in > 50% of the patients. We conclude that pulse dye densitometry using indocyanine green 5 mg or 20 mg is inaccurate in anaesthetised patients before coronary artery bypass surgery and cannot be used after surgery because of a high incidence of low pulse dye densitometry signal amplitudes.