Observer variation in blood pressure measurement following training with standard techniques has been investigated in a study of 7735 middle-aged men. The initial training produced consistent results between observers, but there were progressive deteriorations in the ensuing months. Subsequent re-training led to marked though inconsistent reductions in variation, but within 1 or 2 months, observers reverted back to an individual level of bias. The magnitude of observer variation is in line with the few other published accounts, and is sufficient to have important implications for group comparisons, clinical trials, and the clinical care of individual patients. Four methods of reducing observer variation are considered: regular re-training, self-measurement and ambulatory monitoring, automatic sphygmomanometers, and adjustment in the analysis. The adjustment procedure used in the British Regional Heart Study is described. Regular and frequent re-training with monitoring of performance, or the use of automatic machines are presently considered the two most practical methods of reducing observer variation.