Summary Environmental managers are often confronted with unplanned or accidental disturbances that may lead to environmental impacts. Procedures for detecting or measuring the size of such impacts are complicated because of the lack of data available before the disturbance and because of the intrinsic variability of most natural measures. Here, a protocol for detecting impacts is illustrated for single-measure variables (numbers of individual species) and multivariate measures (relative abundances of invertebrates in assemblages). The present paper describes a case concerning drainage of acidified water into an estuary due to construction of a drainage channel in an area of wetland for which there had been no prior investigations (i.e. no ‘before’ data). The spatial extent of any impact was also unknowable. Sampling was, therefore, designed to allow for impacts of only a few tens of metres (using control sites 50 m from the mouth of the channel) and impacts covering much larger areas (500 m and 1 km from the mouth of the channel). Invertebrates in the mud around the channel and in control sites were sampled in replicated cores and the amount of seagrass in each core was weighed. Average abundances of invertebrate animals and weights of seagrass were compared, as was variation among samples in potentially impacted and control sites (using univariate analyses of variance). Sets of species were compared using multivariate methods to test the hypothesis that there was an impact at one of the scales examined. In fact, there was no evidence for any sort of impact on the fauna or seagrasses; the disturbance was a short-term pulse without any obvious or sustained ecological response. One consequence of the study was that the local council was able to demonstrate no impact requiring remediation and no penalties were imposed for the unapproved construction of the channel. The implications of this type of study after an environmental disturbance are discussed. The present study identifies the need for clear definition of relevant hypotheses, coupled with rigorous planning of sampling and analyses, so that reliable answers are available to regulators and managers.