Doppler cardiac output (CO) determination is discussed as a non-invasive alternative to CO estimation by thermodilution. This study was designed to compare the accuracy of a new transoesophageal Doppler device with the thermodilution technique. In 24 patients undergoing coronary artery bypass surgery, CO was determined simultaneously by the oesophageal Doppler (OD) and thermodilution (TD) method in triplicate for three sample episodes: after induction of anaesthesia during clinical steady-state conditions (A), after start of surgery (B), and after sternotomy (C). The agreement between ODCO and TDCO estimations was assessed by analysing the mean difference, indicating the systematic error, and analysing the distribution of differences between the two methods. The bias between ODCO and TDCO estimations was 0.38 (−0.06 to +0.81) L min−1 (mean and 95% confidence interval) for sample episode A, 0.48 (−0.11 to +1.1) L min−1 for sample episode B, and 0.69 (+0.08 to +1.3) L min−1 (P < 0.05 vs. zero) for sample episode C. Bias analysis of the log-transformed data revealed that 95% of the ODCO values differed from TDCO values by 43% below to 50% above for sample episode A, by 39% below to 95% above for sample episode B, and by 32% below to 96% above for sample episode C. Analysis of the changes in CO from sample episode A to B and from sample episode B to C, expressed as percentage values, showed a non-significant bias between the methods, but the 2 SD limits were ±44% and ±36% respectively. Our findings suggest that CO estimation by OD cannot replace estimation by the TD method.