To project mean weekly stream temperature changes in response to global climate warming and for studies of freshwater ecosystems, a four-parameter nonlinear function of weekly air temperatures was used. One parameter, the upper bound stream temperature, was obtained by extreme value analysis from stream temperature data, and the other three parameters were obtained by least squares regression analysis. The least squares regression function was developed separately for the warming season and the cooling season (hysteresis) to take heat storage due to snowmelt or reservoir operations into account. There were very weak correlations between model parameters and annual or seasonal air temperatures. To project weekly stream temperatures under a 2¿CO2 climate scenario, weekly air temperature data from 166 weather stations, incremented by the output of the Canadian Center of Climate Modelling (CCC) general circulation model (GCM), were applied to nonlinear stream temperature models developed for 803 stream gaging stations. An error analysis indicated that only 39 stream gaging stations would not exhibit a significant change under the CCC-GCM 2¿CO2 climate scenario. The projections at the remaining 764 stream gaging stations showed that mean annual stream temperatures in the contiguous United States would increase by 2¿--5 ¿C, least near the West Coast and most in the Missouri River and Ohio River basins. On average, there would be a 1¿--3 ¿C increase in the maximum and minimum weekly stream temperatures under the 2¿CO2 climate scenario, most in the central United States. It was also found that most streams would experience the maximum change in weekly stream temperatures in spring (March--June). The minimum changes in stream temperatures are projected to occur in winter (December and January) and summer (July and August) throughout the United States. ¿ 1999 American Geophysical Union