Temporal variability in water temperature plays an important role in aquatic ecosystems, yet the thermal regime of streams has mainly been described in terms of mean or extreme conditions. In this study, annual and diel variability in stream water temperature was described at 135 unregulated, gauged streams across the USA. Based on magnitude, amplitude and timing characteristics of daily water temperature records ranging from 5 to 33 years, we classified thermal regimes into six distinct types. This classification underlined the importance of including characteristics of variability (amplitude and timing) in addition to aspects of magnitude to discriminate thermal regimes at the continental scale. We used a classification tree to predict thermal regime membership of the six classes and found that the annual mean and range in the long-term air temperature average along with spring flows were important variables defining the thermal regime types at the continental scale. This research provides a framework for a comprehensive characterization of the thermal regimes of streams that could provide a basis for future assessment of changes in water temperature caused by anthropogenic activities such as dams, land use changes and climate change.