To investigate the accuracy of three non-invasive blood pressure (BP) devices in atrial fibrillation (AF) compared with invasive arterial BP. One hundred patients aged 45-90years, 63% male (50 in AF and 50 age matched controls in sinus rhythm [SR]) were identified with arterial lines measuring beat-to-beat BP fluctuation. Non-invasive BP measurements utilising the manual sphygmomanometer (MS), PulseCor R6.5 (PC) and automated sphygmomanometer (AS) were taken simultaneously with invasive BP in a randomised sequence. This was repeated three times in each patient. In SR differences in systolic BP (SBP) for MS, AS and PC were -0.34mmHg (95% CI -2.31 to 1.63; P=.733), -3.80mmHg (95% CI -5.73 to -1.87; P=.0001) and -3.90mmHg (95% CI -5.90 to -1.90; P=.0001) and for diastolic BP (DBP) were 6.02mmHg (95% CI 4.39-7.64; P<.0001), 8.95mmHg (95% CI 7.36-10.55; P<.0001) and 7.54mmHg (95% CI 5.89-9.18; P<.0001), respectively. In AF mean differences in SBP for MS, AS and PC were -7.33mmHg (95% CI -9.11 to -5.55; P<.0001), -5.29mmHg (95% CI -7.08 to -3.50; P<.0001) and -5.75mmHg (95% CI -7.54 to -3.96; P<.0001) respectively and for DBP were 5.28mmHg (95% CI 4.03-6.54; P<.0001), 6.26mmHg (95% CI 5.00-7.52; P<.0001) and 6.89mmHg (95% CI 5.64-8.15; P<.0001) respectively. The MS is accurate in SR because of direct assessment of Korotkoff sounds. Non-invasive BP assessment in AF is significantly less accurate. These findings have important prognostic and therapeutic implications.