The accuracy of self-monitoring of blood glucose (SBGM) was assessed by a quality control programme. Ninety diabetic patients who were routinely performing SBGM were supplied with a series of quality control solutions which they tested with their usual meter and reagent strip. The overall error rate (a result outside the range of the mean +/- 3SD for each quality control solution) was 39% for users of the Ames system and 33% in Boehringer Mannheim system users. The clinical relevance of these errors was determined by examining the effect on the clinical decision which would have been based on the erroneous result, either in taking inappropriate action or in failing to take appropriate action. In 30% of all patients, 25% or more of the errors were of such a degree as to be clinically misleading. The most common error was an underestimation of the result, which gave the impression of better than actual blood glucose control. The main reason for these errors was the failure of patients to take sufficient care in following the manufacturers' instructions when performing the test. Clinically relevant erroneous results are common among patients performing SBGM.