OBJECTIVES: The reporting of blood pressure (BP) is known to be affected by errors due to digit preferences or rounding to the closest multiple of ten. If this occurs completely randomly, estimates of effect will be attenuated and the power of statistical tests reduced. Systematic bias can also occur if, for instance, physicians round measurements differentially for high or low risk patients. We examined this with data taken from an ongoing naturalistic study of the impact of a hypertension control education program. METHODS: Systolic and diastolic blood pressure measurements were reported by physicians at the time of enrolling hypertensive patients. We derived the distribution of the terminal digit of the measurements and compared the probability of reporting a terminal zero when the measurement was above or below the threshold of BP control (140/90 mm Hg) and by patients' baseline characteristics. RESULTS: Analyses were based on an interim cut of data on 4594 patients, of which 47.5% were male with mean age 56 years (SD=13.5). There was clear evidence of digit preference by physicians: e.g., 46.3% of systolic BP measurements ended with 0, and 43.3% were even numbers. The probability of a terminal zero was slightly higher above the systolic BP control threshold: 47.1% above vs. 44.4% below (p-value=0.09). Rounded values were also more likely for females (47.9 vs. 44.6%, p-value=0.02) and patients 65 years or older (49.7 vs. 45.2%, p=0.01), but did not differ by levels of other cardiovascular risk factors. Measurements of diastolic BP were also rounded very frequently (44.6%), but differences by patient characteristics were not significant. CONCLUSION: Physicians frequently round to the closest multiple of ten when reporting BP measurements, and do so slightly more frequently for female and older patients, and those further from the threshold defining BP control.