We performed a two-stage prospective evaluation of the error detection capabilities of duplicate analysis of blood-gas specimens. In the first stage we analyzed 1601 specimens with a Corning Model 175 blood-gas analyzer as the test instrument and a Corning Model 178 analyzer as the reference instrument, and in the second stage we analyzed 1544 specimens with two Model 178 analyzers. In each stage the designated reference instrument underwent troubleshooting whenever an analytical error was detected; the test instrument underwent troubleshooting only when error conditions were indicated by means other than duplicate analysis. An error was considered to have occurred if the difference between the duplicate analyses exceeded 0.02 (for pH), 0.53 kPa, i.e., 4 mmHg (pCO2), or 7% (pO2). The number of specimens for which errors were detected was 97 (6.1%) in the first stage, 23 (1.5%) in the second. For each analyte more errors were detected with the Model 175 analyzer (of older design) than with the newer Model 178 analyzer. Furthermore, in certain periods associated with the use of particular electrodes there were very high error rates for individual analytes: 8% for pCO2, 18% for pO2. We conclude that duplicate analysis should be considered as a possible required standard for error detection.