Our understanding of reef dynamics is based on our ability to detect biological differences, or effects, in time and space. Most studies compare variability across scales: they test signal-to- noise ratios. When effects cannot be detected, the noise (error variance) may be obscuring trends at larger or smaller scales. Conversely, effects detected at one scale may be meaningless at other scales. As an example, Acropora cervicornis was recently eradicated by white-band disease and replaced by Agaricia tenuifolia on lagoonal reefs in Belize. This transition occurred over a large area and was clearly a signal on a decadal scale. There was no way to tell from ecological observations, however, if it constituted a signal at larger time scales; such transitions could have been common in the past, obscuring millennial-scale trends. Coring studies revealed that reefs of the central lagoon were continuously dominated by A. cervicornis for at least the last 3,000 years. Previous Acropora-to-Agaricia transitions were apparent in two cores, but these singnlar events dated more than 1,000 years apart. There was no evidence of an earlier transition as spatially widespread as the recent one. Geographically, then, the recent transition was a signal, whereas the two earlier transitions were local noise. Temporily, the transition was unique on a millennial scale, and its association with increased human activity suggests that it was a signal rather than noise. Formulating a hypothesis at the right scale is even more important than selecting the best sampling program for the problem.