The use of BACI (before-after, control-impact), comparisons to assess environmental impacts on communities requires strict adherence to the principles of good experimental design in order to derive unambiguous conclusions. Unplanned natural ecological experiments seldom meet these requirements. What unplanned experiments lack in rigor, they often make up for in an abundance of data that has the potential to make even weak signals detectable. To realize this potential the following conditions should be satisfied.(1) High taxonomic resolution is achieved in identification (species level if possible).(2) Species-specific tolerance values are available for most of the collected taxa.(3) An individual measure of impact is obtained for each taxon at a site. (4) All taxa are used in impact assessment. As a practical illustration we examine the effect of sedimentation from construction runoff on the population levels of stream benthic macroinvertebrates. Our purpose is to assess the efficacy of erosion-and sedimentation-control regulations and their enforcement in three North Carolina (USA) jurisdictions. Due to the asynchronous onset and duration of construction events, samples from different sites varied both in their timing and frequency. As a result, the benthos (both the kinds and numbers of taxa collected) overlapped so little between sites that the standard multimetric approaches to impact assessment were of little use. The typical multivariate methods of impact assessment examine changes at the community level. Our approach focuses on changes at the individual species level for all collected taxa. By correlating species-level population changes with known species-level tolerance values we are able to construct an impact score for each site along with a measure of its reliability. Impact scores are then used to test for jurisdiction differences using a weighted regression approach to analysis of variance. If the immediate goal of a study is not to assess the impact at a single site but to compare the degree of impact across a number of impacted sites and to relate this to one or more classification variables, then we believe this approach can be broadly applicable.