High-quality calibration data sets are required when diatom assemblages are used for monitoring ecological change or reconstructing palaeo-environments. The quality of such data sets can be validated, in addition to other criteria, by the percentage of significant unimodal species responses as a measure of the length of an environmental gradient. This study presents diatom-environment relationships analyzed from a robust data set of diatom communities living on submerged stones along a 2,000 km long coastline in the Baltic Sea area, including 524 samples taken at 135 sites and covering a salinity gradient from 0.4 to 11.4. Altogether, 487 diatom taxa belonging to 102 genera were recorded. Detrended canonical correspondence analysis showed that salinity was the overriding environmental factor regulating diatom community composition, while exposure to wave action and nutrient concentrations were of secondary importance. Modeling the abundances of the 58 most common diatom taxa yielded significant relationships with salinity for 57 taxa. Twenty-three taxa showing monotonic responses were species with optimum distributions in freshwater or marine waters. Thirty-four taxa showing unimodal responses were brackish-water species with maximum distributions at different salinities. Separate analyses for small (cell biovolume <1,000 μm(3) ) and large (≥1,000 μm(3) ) taxa yielded similar results. In previous studies along shorter salinity gradients, large and small epilithic diatom taxa responded differently. From our large data, we conclude that counts of large diatom taxa alone seem sufficient for indicating salinity changes in coastal environments with high precision.