Two different methods of assessing the reliability of the oxygen electrode of one model of an automatic blood gas analyser (BGA) have been studied. In the first, a single automatic BGA was assessed by using outdated bank blood which was pumped around a small extracorporeal circuit into which known gas mixtures were passed. Oxygen tension was varied between 2 and 16 kPa. In the second, fresh heparinized blood was tonometered with calibrated gases and submitted to the automatic BGA used in the first part of the study and also to three other identical machines. Each of the machines was between 3 and 4 years old.Eighteen different units of blood were used in the first part of the study. The correlation coefficient between the automatic BGA and the Po(2) in the extracorporeal circuit varied between 0.29 and 0.99. 31% of the total of 209 measurements made by the automatic BGA were more than 1.2 kPa from the reference value, 25% of them being between 1.2 and 4.0 kPa from the reference value. In the second part of the study, the correlation coefficient between this automatic BGA and the tonometered blood was 0.96. The correlation coefficients for the 3 other identical BGAs were 0.84, 0.97 and 0.88, indicating that the BGA used in the first part of the study was no worse than any of the others.It is suggested that although clinicians are likely to ignore readings of an automatic BGA that are more than 4.0 kPa from the true value and are likely to repeat the investigation, readings between 1.2 and 4.0 kPa from the true value may adversely affect patient management.