There is a variety of methods used for monitoring environmental variables for the detection of ‘unacceptable' levels, such as for compliance conditions. In this paper, we consider the utility of a statistical technique used for many years in industrial process controlcumulative sum or CUSUMin monitoring water-quality trends within large storages such as reservoirs. By using simulations and realistic values for total phosphorus concentrations, we show the sensitivity of CUSUM schemes to six kinds of trends in quality through time. The six trends are as follows: (a) on target; (b) off target; (c) midpoint (halfway between a and b); (d) ramping change (increasing from a to b over 5 years); (e) weakly autocorrelated series; and (f) strongly autocorrelated series. We study the performance of the CUSUM method in relation to ‘average run lengths' (ARLs) before a warning is triggered and the probabilities of spurious warnings (type-I statistical errors) or failing to detect changes in water quality that are deemed to be unsatisfactory. We conclude that in reasonably buffered systems, such as large storages, the CUSUM technique may be a potentially important way of monitoring water quality that also permits rapid response to serious increases in nutrient concentrations.