Thermodilution cardiac output measurements are commonly employed in the management of critically ill patients. Serial measurements often show significant variation, and poor reproducibility limits their clinical utility. There are no clinical studies revealing when to perform thermodilution cardiac output measurements in relation to the respiratory cycle. We prospectively studied 32 patients in a randomized scheme comparing three thermodilution cardiac output measurements at peak-inspiration, at end-exhalation, or randomly in spontaneously breathing and mechanically ventilated patients. Saline injections initiated at peak-inspiration or end-exhalation resulted in cardiac output measurements with much smaller standard deviations than those seen with random injections. Thermodilution cardiac output measurements performed at random times in the respiratory cycle should be avoided, and we recommend initiating these measurements at end-exhalation.