1. Benthic diatoms have been widely used as indicators of water quality in streams. New insights that diatoms may also respond to large-scale drivers, such as climate or historical factors, highlight the need to reassess the usefulness and the reliability of diatoms as bioindicators. 2. Using a suite of modelling techniques, weighted averaging (WA), weighted averaging partial least squares, modern-analogue technique (MAT) and two machine learning techniques, boosted regression trees (BRT) and random forests (RF), we calibrated models to infer water quality and climatic variables using diatom abundance data collected from 227 stream sites in Finland. 3. Predictive ability was generally better for climatic variables [growing degree days (GDD) defined as temperature >5°C, summer precipitation and water balance] than for local environmental variables (conductivity, water colour and total phosphorus). The strongest relationships were found for GDD (r 2 = 0.86, MAT) and conductivity (r 2 = 0.82, RF). Using BRT, we also identified potential indicator species for local environmental and climatic variables, based on relative importance species in the models. 4. Our results show that diatoms could serve as efficient proxies for climatic variables and local environmental conditions. Furthermore, new modelling techniques such as modern regression trees can provide new insights into relationships between diatom assemblages and local water quality and climate, and thus help to construct more reliable indices. These methods could also serve as important tools to infer environmental variables in changing ecosystems.