To investigate the ideal sampling interval for the estimation of heart rate (HR) at rest and during exercise in atrial fibrillation (AF), maximal exercise testing with continuous electrocardiographic acquisition was performed in 10 patients with chronic AF (mean age 66 ± 4 years) and 10 subjects in normal sinus rhythm (mean age 31 ± 6 years). Measurements of HR were obtained at 9 different sampling intervals (1, 2, 3, 6, 10, 15, 20, 30 and 60 seconds) at rest and 7 different sampling intervals (1, 2, 3, 6, 10, 15 and 20 seconds) during the last 30 seconds of each minute during exercise. The HR obtained from each interval was compared with true HR (determined by a 4-minute sample at rest and by the last 30 seconds of each minute during exercise). Among patients with AF, large differences were observed between the HR obtained and true HR, both at rest and during exercise, using small sampling intervals. The mean of these differences ranged between 16 ± 11 beats/min (range 14 to 22) using 1-second sampling intervals and 2.2 ± 2.0 beats/min (range 1.6 to 4.4) using 20-second sampling intervals during progressive exercise. Variability of the HR obtained from a given random sample was also high when short sampling intervals were used among patients with AF. These observations were contrasted by subjects in normal sinus rhythm, among whom neither variability nor measurement error were influenced remarkably by changing the sampling interval or increasing HR.