Due to the sparse data on the repeatability of short and ultra-short term heart rate variability (HRV) measures, we measured the repeatability of common HRV measures derived from 10-second, 2-minute, and 6-minute recordings in 63 healthy men and women, aged 45-64, in Chapel Hill, North Carolina. Three 10-second and 2 six-minute heart rate recordings were obtained during each of 2 visits, separated by 1 to 2 weeks. We partitioned the measurement error into components and computed intraclass correlation coefficients using nested, random effects models. Repeatability improved with the length of recording: intraclass correlation coefficients were greater than 0.7 for 6-minute measures and 2-minute time domain measures and greater than 0.5 for 2-minute frequency domain measures. Repeatability of measures from 10-second records was lower, but improved considerably when the mean from 2 or 3 records was used. Correlations between the same measures from different length recordings were quite high. Our findings support the use of records of at least 5 minutes in length in epidemiological studies, in accordance with previous guidelines. Researchers using 10-second records should consider taking the mean of several recordings, when possible, or using statistical methods to correct for measurement error.