In community monitoring an attempt is made to identify long-term trends by regularly sampling of selected sites. Since the benthos is reputed to be fairly sedentary, the spatial resolution is often reduced to single sites. However, members of many benthic invertebrate species have been found drifting across sedimentary seabeds in shallow waters. Transportation by currents may result in changes of their spatial pattern in the sediment, thereby changing local community composition. The quantitative importance of drifting was tested by repeated sampling of a 2-km² shallow (10 m) offshore area west of the island of Sylt (North Sea). Within the fortnight period between two samplings the benthic community composition had changed dramatically. Despite fairly calm weather, translocation of organisms by currents exceeded 1 km. In about half of the species, the spatial changes in abundance within these two weeks roughly equalled the average variation between consecutive years. This example suggests that community monitoring needs a wide spatial scale to discriminate long-term temporal changes from short-term variability. Extending the sampling area from 2 to 180 km² strongly reduced the variability of abundance estimates. However, only in a few species was the spatial distribution over the sampling sites found during one sampling date a suitable estimator for the spatial pattern found one or two months later, at the same sites. Instead the spatial patterns of the fauna changed strongly during a single month with a spatial scale of re-distribution exceeding several km in some species. At the same time the granulometric sediment composition changed, indicating changes of habitat quality. Hence, sampling of a large area, with random selection of the sampling sites on each sampling date, is suggested to yield the most reliable estimates of population development in the coastal North Sea. However, in view of the expected spatial scale of re-distribution during storm tides and the spatial variability of recruitment, even a 180-km² sampling area may be too small.