OBJECTIVES: Clinicians have a strong tendency to round blood pressure (BP) readings, particularly to the closest multiple of ten, even when using precise digital devices. This poses a challenge for classifying patients as controlled in economic evaluations when values fall on the boundary of the definition (e.g., 140/90mmHg), since including or excluding the boundary values will over- or under-estimate control rates. We describe a simple sensitivity analysis to gauge the impact of rounded readings. METHODS: The method attempts to correct the excess number of values ending in zero by redistributing these in the ±5mmHg range around the observed value. Thus, a value of 140 mmHg is replaced by a new hypothetical measurement drawn randomly between 135-144 mmHg. The correction is made for 90% of observations falling on the boundary since 10% would naturally be expected to end in zero. We illustrate the method with simulated data and validate by comparing the proportion below the threshold with the true, the rounded and the corrected BPs. The correction was applied to data from an ongoing study of a hypertension control education program. RESULTS: A sample of 1000 random systolic BPs was generated with mean 155 and variance 30; these “true” data were then distorted by rounding 40% of values (as observed in the study). The 30.8% in the true data that were controlled (<140mmHg) dropped to 28.6% after rounding; after the correction, the proportion was 30.9%. In time-to-control (<140/90mmHg) analyses of the study data, 34.2% of patients had controlled BP at 6 months based on the observed data. Replicating the analyses with corrected BPs yielded an estimate of 41.9%. CONCLUSION: The impact of rounding should be taken into account in analyses of BP data to minimize bias in control rates, as well as attenuation of treatment effect estimates that might result.