AbstractIn this paper, the Bayesian model and Lepage test were used to detect change point and to analyse associated statistical properties of high/low water levels in summer (June, July and August (JJA)) and winter (December, January and February (DJF)) months across the PRD (Pearl River Delta). The results indicate that: (1) two time intervals, that is 1979–1981 and 1988–1995, witness abrupt changes of SmH/SmL (summer mean high water level/summer mean low water level). The lower PRD is dominated by increased mean and coefficient of variation (Cv) of SmH. Increased mean but decreased Cv of SmL can be observed in the Mainstem Pearl River; (2) WmL (winter mean low water level) and WmH (winter mean high water level) of about 74% of the total stations have two change points occurred roughly during 1969–1971 and 1993–1995. First change points of WmH are mainly characterized by increased mean and Cv, but decreased mean and increased Cv of WmL can be observed across major parts of the PRD. The driving factors causing abrupt changes of water levels are various. Intensive human activities cannot be ignored, for example in‐channel dredging and reallocation of the streamflow within the river channels due to human‐induced topographical changes of river channel. Different responses of high/low water levels to externally influencing factors and interactions between influencing factors make the alterations of the water levels across the PRD more complicated. The findings of this paper will be helpful for the management of the PRD and human mitigation to natural hazards under the changing environment. Copyright © 2008 John Wiley & Sons, Ltd.