THE prospect of global warming has focused attention on the role of palaeoecology in testing the accuracy and sensitivity of climate-model predictions, in identifying past analogues for future climate change, and in placing model-predicted climate responses in the context of natural climate variability1,2. Proxy data for climate reconstruction can be derived from many sources, including the palaeolimnological record3,4. In closed-basin lakes in arid and semi-arid regions, shifts in effective moisture lead to the concentration or dilution of dissolved salts, and these changes in salinity are clearly reflected in the composition of lacustrine diatom assemblages5-8. Here we refine a previously published9 diatom-based transfer function for the reconstruction of past changes in salinity of lakes in the northern Great Plains region of North America, and apply the refined transfer function to a late-glacial and Holocene sediment record from Devils Lake, North Dakota. Our results show that there were a number of alternations between fresh and saline conditions during the Holocene and hence demonstrate the utility of the technique in reconstructing past changes in regional climate.