OBJECTIVE The objectives of our study were (1) to compare mixed venous saturations calculated by a blood gas machine with those measured directly by a co-oximeter; and (2) to compare the sensitivities and specificities of VO2s derived from these values. METHODS Charts were retrospectively reviewed of all MICU patients [n = 16] between December 1, 1991 and January 31, 1992, who required pulmonary artery catheters for their usual care and who had hemoglobin saturations of mixed venous blood concurrently measured by both a co-oximeter (Co-Ox Model 482, Instrumentation Lab, Lexington, MA) and a blood gas analyzer (Nova Biomedical StatLab5, Waltham, MA) which uses a variant of the Severinghaus equation to calculate SVO2 from PVO2). Data used at the time of each SVO2 measurement to calculate oxygen consumption (VO2) further was collected. RESULTS Available for analysis were 118 mixed venous blood samples. Although the SVO2 values had a correlation coefficient of 0.807 (95% confidence interval [CI] 0.736 to 0.861, Fisher's z-transform), when VO2s were calculated, the blood gas analyzer calculated saturations had a sensitivity of only 58.3% and a specificity of 89%, when compared with those calculated using the saturations measured by the co-oximeter. Attempts to mathematically improve upon the Severinghaus equation and upon an additional four regression equations used by other blood gas analyzers resulted in universally worse sensitivity. CONCLUSIONS If SVO2s calculated by a blood gas machine--rather than those co-oximetrically measured--are used to calculate VO2s, 42% of patients with low O2s will be misclassified as normal and 11% of normals will be misclassified as low. This total error appears to be the result of measurement error by the PO2 electrode of the blood gas analyzer and shifts of the oxyhemoglobin dissociation curve, which are not accounted for in the equation that is used to calculate saturation from measured PO2. We were not able to improve mathematically the sensitivity of any of the available regression equations used by blood gas analyzers to calculate SVO2 from PVO2. Therefore, it remains necessary to use co-oximetrically measured saturations when calculating VO2.