OBJECTIVES: (1) To compare blood pressure (BP) readings with an automated arm cuff oscillometric device (AutoBP) to readings with a mercury sphygmomanometer (HgBP) and (2) to evaluate the impact on the prevalence of hypertension (HBP) in a population-based survey. METHODS: (1) In a convenience sample ("Comparison Study"), we measured BP with both AutoBP (Visomat(R) OZ2) and HgBP and we modeled BP difference (DeltaBP = HgBP-AutoBP) with multiple regression analysis. (2) Using DeltaBP, we calculated HgBP in a survey previously conducted in Dar es Salaam ("Population Survey") in which BP was measured with the automatic device Visomat(R) OZ2 and we compared the prevalence of HBP (>/=140/90 mmHg or treatment). RESULTS: In the Comparison Study (404 subjects aged 25-64), systolic/diastolic BP was higher by 4.4/4.7 mmHg (SE: 0.4/0.3) with HgBP than AutoBP. The prevalence of HBP was 42% with HgBP and 36% with AutoBP (relative difference of 14%). DeltaBP was associated with age, BP and arm circumference. In the Population Survey (9.254 subjects aged 25-64), the prevalence of HBP was 17% with calculated HgBP and 14% with AutoBP (relative difference of 20%). CONCLUSION: A small systematic bias in BP readings between two different devices had large impact on hypertension prevalence estimates. This suggests that automated devices used in epidemiological studies should be validated with particular care.