Thirty-one critically ill patients with acute respiratory failure and circulatory shock were divided into three groups. Group 1 included 11 patients with an inserted mixed venous saturation (SvO2) catheter using three-reference wavelengths and characterized by one transmitting and one detecting fiberoptic filament (Oximetrix opticath catheter); group 2 included eight patients with an inserted SvO2 catheter using two-reference wavelengths and characterized by one transmitting and one detecting fiberoptic filament; (Edwards sat-one catheter); group 3 included 12 patients with an inserted SvO2 catheter using two-reference wavelengths and characterized by one transmitting and two detecting fiberoptic filaments. Once calibration procedures were performed, SvO2 measured by the catheter and by an hemoximeter OSM 3 (reference value) were compared following each therapeutic intervention. Over a period of 1.5 to 6 hours during which the hematocrit value remained unchanged, 119 measurements were obtained in group 1, 91 in group 2 and 181 in group 3. The dispersion of SvO2 values was much more pronounced with the two-reference wavelength systems using either one or two detecting fiber optic filaments, and the correlation coefficient was significantly higher with the three-reference wavelength system (r = 0.970 for the Oximetrix catheter vs r = 0.855 for the Edwards catheter and r = 0.826 for the Spectramed catheter, p less than 0.001). After 24 hours, the spontaneous drifts in the two-reference wavelength systems, using either one or two detecting fiber optic filaments (expressed as the SvO2 value measured by the catheter minus the reference SvO2 value) were significantly higher than the spontaneous drift in the three-reference wavelength system (9.3 +/- 7 percent for the Edwards catheter and +/- 6 +/- 4.1 percent for the Spectramed catheter vs 3.3 +/- 3.1 percent for the Oximetrix catheter, p less than 0.05). This study shows that a three-wavelength system is more accurate than a two-wavelength system for measuring acute changes in SvO2. The addition of a second detecting fiber optic filament does not seem to improve the accuracy of the system when SvO2 changes occur in conditions of stable hematocrit.