Background: Bolus thermodilution cardiac output measurements have been a mainstay in clinical monitoring of critically ill patients for more than 30 years. Usually the results of an arbitrarily chosen number (1-6) of thermal indicator injections are averaged to increase the reliability of the measurement. The number of injections needed to achieve a given level of precision has, however, not previously been systematically investigated. Methods and results: In 80 hemodynamically stable patients cardiac output was determined as the average of eight injections of 10 ml of iced saline. From the 638 measurements we examined the relationship between the number of thermal indicator injections and the precision of the resulting cardiac output estimate. Furthermore, the association between the number of injections and the least detectable difference among two sets of measurements was established. Conclusion: The current study shows that one needs to average the results of four injections to be 95% confident that the result is within 5% of the 'true' cardiac output and that two series of four measurements have to differ by at least 7% before one can be sure (95%) that a change in cardiac function has taken place.