Despite the fractal properties of the linkages between catchments and rivers, the tendency in almost all the work on land use and nutrient loads to receiving waters has been to use large-scale and long-term averages of water quality data collected in rivers and streams to assess downstream impacts. This is partly because of two fundamental biases and misconceptions: first, that the usual regime of weekly or more infrequent sampling is sufficient, and second, that the observed variability in the data is just noise, which can be averaged and dealt with statistically. Recent analysis of high-frequency data from rivers, lakes and coastal waters (and by high-frequency I mean data collected daily or even more frequently than that – down to minutes in some cases) indicates that both these basic assumptions are false. The data are not just ‘noisy’, there is information there, and weekly or less frequent sampling does not resolve the true scales of pattern and process. In short, catchments have fractal flow paths and these produce fractal distributions of patches of water flowing down rivers and into receiving waters. The species that characterise both terrestrial and aquatic ecosystems interact to both produce and exploit these fractal patterns; they are part and parcel of the reaction–diffusion relationships between organisms at small scales.