The use of a pulse oximeter to monitor arterial oxygen saturation (SaO2) is considered accurate and reliable in the range of 90% to 100%. However, differing reports exist about the accuracy with desaturation. Thus, the suitability of pulse oximetry in desaturated patients was evaluated using a Nellcor N-100 oximeter. In 56 children with cyanotic congenital heart disease, the pulse oximeter reading was compared with the direct measurement of SaO2 by a CO-oximeter OSM 3. The influence of high hematocrit values on the accuracy at low saturation was also investigated. All oxygen saturation measurements (two per child) were carried out after induction of anesthesia (ketamine, fentanyl, pancuronium) during a "steady state" before the surgical procedure. The results indicate that at low levels of saturation (SaO2 below 80%), pulse oximetry is not as accurate as at higher saturations, and overestimates the true value. Bias and precision between saturations measured by the pulse oximeter and the CO-oximeter were 5.8 and 4.8 in the group with a saturation below 80%, and 0.5 and 2.5 in the group with a saturation over 90%, respectively. Because the margin of safety for a patient is small when arterial saturation levels are under 80%, it is advisable under this condition to check the SaO2 measurements by a CO-oximeter. High hematocrit levels did not seem to be responsible for impaired accuracy of pulse oximetry at saturation values below 80%.