We have conducted a voluntary, community blood-gas proficiency testing program, with use of tonometered human blood, for 32 analyzers located in 16 laboratories. Instruments initially showed inaccuracies as large as -30.8 to +17.3% for po(2), and -14.0 to +42.9% for pco(2), but inaccuracy and imprecision decreased in most laboratories during the program. For a typical 15-week period, mean group precision (CV) was 4.3 To 5.1% for po(2) from 6.92 to 33.3 Pa (52 to 250 mmHg), and 4.0 to 6.9% for pco(2) from 2.0 to 6.8 Pa (15 to 51 mmHg). This program can detect increasing imprecision or inaccuracy caused by analyzer deterioration, and can identify interlaboratory or interinstrument bias and problems not detected by participant quality-control programs. Participants have used the proficiency information in discussing data quality with clinicians, promoting internal control and maintenance programs, and justifying instrument purchases. We believe that proficiency-testing documentation of variability in blood-gas analysis may help to establish realistic patient-care protocols.