The association of diatom assemblages to salinity was studied in 95 lakes and streams ranging from freshwater to hypersaline in the south-west of Western Australia. The relationship between environmental variables and species composition was explored using canonical correspondence analysis (CCA) and partial CCA. Salinity was shown to account for a significant and independent amount of variation in the diatom data, enabling a transfer function to be developed based on the final dataset, which consisted of 89 sites and 150 diatom taxa. The most successful model was derived using tolerancedownweighted weighted averaging. Summary statistics showed that the transfer function performed very well with a high coefficient of determination and low prediction errors that remained high after the crossvalidation method of jackknifing (r apparent 2 = 0.97 and r jackknifed 2 = 0.89). This suggests that salinity can be accurately predicted using relative abundances of diatoms, and the model can now be applied to paleolimnological reconstructions. However, the transfer function also provides the basis for use in future biomonitoring studies to detect increases in salinity for lakes and streams most at risk, as well as to evaluate the success of remediation measures implemented to secondary salinised systems.