1. Relatively undisturbed streams in continental U.S.A. were classified according to variation in ten ecologically relevant hydrological characteristics. Measures of flow variability and predictability for average conditions, as well as for low- and high-flow extremes, were extracted from long-term (15-58-year) daily streamflow data for 806 streams. 2. Two groups of sites were analysed: all 806 sites and a subset of 420 'best' sites. For each group, cluster analysis identified ten distinctive stream types, seven permanent and three intermittent. The geographical clustering exhibited by the stream types indicated regional differences in climatic and geological features. A bootstrapping technique applied to the permanent stream classes showed the majority of them were statistically robust. 3. The derived classification of U.S. streams based on ecologically relevant hydrological characteristics provides a comprehensive catalogue of small to mid-size streams that, according to ecological theory, may differ in major aspects of their ecological organization. The classification provides a basis for hypothesis generation and affords an objective framework for matching streams for purposes of comparative ecological investigations. 4. A subset of 118 streams from the ten classification groups was selected to determine whether certain hydrological variables often used by ecologists to make cross-system comparisons are sensitive to the temporal coarseness of the hydrological time series used to derive the variables. The three hydrological variables considered were streamflow predictability, streamflow variability and flood timing. 5. Streamflow predictability (using Colwell's Index) was calculated at daily, weekly, monthly and seasonal time scales. Estimates of predictability showed either no change across time scales, a gradual and consistent increase across time scales, or a maximum value at the monthly time scale. Coefficient of variation of streamflow was calculated at daily, weekly, monthly, seasonal, and annual time steps. Daily values were always greatest for all streams. Some groups showed minimum variability at the monthly scale, others at the seasonal. Timing of daily peak flows could be detected with 50-90% accuracy across stream groups using coarser grain monthly and annual hydrographic data. 6. Inferences about hydrological similarity among streams across broad geographical scales can be sensitive to choice of time scale used in the hydrological characterization.