AbstractThe measurement of changes in the structure of natural marine communities is widely used for the detection and monitoring of man's impact on the sea. Such studies are almost always a compromise between the scientific ideal and political, financial and logistical constraints. This paper considers a number of these constraints, suggests some possible solutions, and discusses the consequences (in terms of loss of information) of adopting strategies which might be considered suboptimal.Identification of all organisms in a community to species level is a major time and cost constraint. Many recent studies have shown that very little information is lost by working at a taxonomic level higher than species (e.g. family or even phylum). This implies a functional coherence of species within each higher taxon. Indeed, there are theoretical reasons, and some empirical evidence, for supposing that community responses to human perturbations may be more easily detected above the noise of natural variability by working at high taxonomic levels.Many environmental impacts, for example oil spillages from tankers, are accidental and it is unusual to have the luxury of a time‐series of data from the site prior to the incident. Good spatial control samples may also be difficult to obtain in a heterogeneous environment. Absolute rather than comparative measures of community degradation are valuable in this context. Methods which fall into this category include comparisons of community structure with some theoretical expectation or with empirical‘training’ data, comparing attributes of community structure that respond differently to perturbation, or identifying highly conservative properties of unperturbed communities that are modified in a predictable way by perturbation.All studies are selective with regard to which components and attributes of the biota should be investigated. The choice frequently depends on local expertise and research interests, rather than on an objective decision about what biological data are most appropriate for the problem in hand. The paper concludes by a discussion of the relative merits of the various alterna