The reliability of cardiac output estimation by thermodilution during artificial ventilation was studied in anesthetized pigs at the right side of the heart. The estimates exhibited a cyclic modulation related to the ventilation. The amplitude of the modulation was independent of the level of positive end-expiratory pressure, ventilatory pattern and volemic loading of the animals. However, a non-constant phase relation existed between the ventilatory cycle and the modulation. Single observations at a fixed moment in the ventilatory cycle are therefore not appropriate for estimation of mean cardiac output nor for studying its relative changes. The averaging of estimates spread equally over the ventilatory cycle led to a much larger reduction in the deviation of the averages from the mean cardiac output than an averaging procedure of randomly selected estimates. The accuracy of estimation of mean cardiac output by two estimates equally spread in the ventilatory cycle was equal to the accuracy obtained by averaging five randomly selected estimates. Averaging four estimates, equally spread in the cycle, appeared to be the optimal procedure. For 89% of all averages an accuracy of 5% around the mean was obtained and for 99% an accuracy of +/- 10%.