Ambulatory blood pressure monitoring (ABPM) allows multiple measurements of BP over a 24h period but often produces a proportion of measurements which are distant from the mean and unlikely to be genuine. The software which accompanies the SpaceLabs 90207 monitor requires criteria to be set which excludes outliers (autoediting) but there is no published data evaluating at what levels these autoediting criteria should be set. This study set out to ascertain whether outlying BP measurements represent true blood pressures or machine error, the determinants of machine error and appropriate autoedit criteria. Twenty subjects, both normotensive and hypertensive, were studied by ABPM with a blinded observer recording BP using a standard mercury sphygmomanometer (SMS) connected via a T-tube during three activity phases: sitting with cuffed arm still, walking but keeping arm still during measurement and sitting but gently moving the arm during measurement. The results show that a high ABPM/SMS discrepancy is associated with arm movement. After studying different autoediting criteria which aimed to eliminate the maximum number of conflicting readings while excluding the minimum number of unconflicting readings, an autoediting threshold of between 1.75 and 2 s.d.s from the mean based separately on sleep and awake measurements is suggested.