Dear Editor-in-Chief: Clemes and Griffiths (2) examined how many days of pedometer monitoring are needed to predict monthly ambulatory activity. They retrace much of the ground covered in a previous report of surprisingly similar title (5). However, certain limitations to their study lead to conclusions that diverge from the earlier findings. Their recording instrument (Digi-Walker SW-200) was less sophisticated than the one that we adopted (modified Kenz Lifecorder). It lacks an acceleration filter to screen out incidental movement artifacts, and because of limited data storage, variance is introduced by frequent removal of the device. Clemes and Griffiths (2) elected as their "gold standard" a 28-d period, despite substantial seasonal variations in physical activity (1,4,5,7). Reliability should be determined for a whole year and not just a single month. The recommendation of a 7-d collection period is particularly questionable because of short-term reactive responses; people who know that they have been fitted with a pedometer walk some 13,000 additional steps during the first week of observation (3). Like us, Clemes and Griffiths (2) have used an intraclass correlation analysis to establish the reliability of data. If a coefficient of 0.8 is accepted, the estimate provides only 64% of the intended information. We used a power spectrum analysis and fast Fourier transformation to evaluate the periodicity of counts over an entire year and defined the number of days of monitoring needed to estimate annual habitual physical activity at specified levels of confidence (5). The necessary period of continuous sampling for individual subjects proved surprisingly long, although obviously it would have been shorter with other approaches to sampling or if the need was simply for averaged information on a large population. In our men, 25 d of consecutive data collection was required to yield a coefficient of 0.8 relative to a yearlong gold standard (5). An individual's movement patterns show a varying vulnerability to exogenous factors such as an adverse climate (1,4,7). Probably because elderly Japanese women usually assume the main burden of low-intensity household tasks (6), the activity pattern of women in our sample was more regular than that of the men, and a coefficient of 0.8 was obtained with 8 d of observation (5). To reach a more satisfactory coefficient of 0.9, 105 and 37 d of consecutive observation were needed in men and women, respectively (5). Finally, Clemes and Griffiths (2) studied a population of working age. Occupations are not specified, but most subjects were presumably used. This would have imposed a structure on weekday activity that is lacking in other segments of the population, such as the elderly people that we studied. Yukitoshi Aoyagi, PhD Exercise Sciences Research Group Tokyo Metropolitan Institute of Gerontology Tokyo, Japan Roy J. Shephard, MD, PhD, DPE, LLD Faculty of Physical Education and Health University of Toronto Toronto, Ontario, Canada