Two experiments were performed to investigate flow beneath an ephemeral stream and to estimate streambed infiltration rates. Discharge and stream-area measurements were used to determine infiltration rates. Stream and subsurface temperatures were used to interpret subsurface flow through variably saturated sediments beneath the stream. Spatial variations in subsurface temperatures suggest that flow beneath the streambed is dependent on the orientation of the stream in the canyon and the layering of the sediments. Streamflow and infiltration rates vary diurnally: Streamflow is lowest in late afternoon when stream temperature is greatest and highest in early morning when stream temperature is least. The lower afternoon streamflow is attributed to increased infiltration rates; evapotranspiration is insufficient to account for the decreased streamflow. The increased infiltration rates are attributed to viscosity effects on hydraulic conductivity from increased stream temperatures. The first set of field data was used to calibrate a two-dimensional variably saturated flow model that includes heat transport. The model was calibrated to (1) temperature fluctuations in the subsurface and (2) infiltration rates determined from measured streamflow losses. The second set of field data was to evaluate the ability to predict infiltration rates on the basis of temperature measurements alone. Results indicate that the variably saturated subsurface flow depends on downcanyon layering of the sediments. They also support the field observations in indicating that diurnal changes in infiltration can be explained by temperature dependence of hydraulic conductivity. Over the range of temperatures and flows monitored, diurnal stream temperature changes can be used to estimate streambed infiltration rates. It is often impractical to maintain equipment for determining infiltration rates by traditional means; however, once a model is calibrated using both infiltration and temperature data, only relatively inexpensive temperature monitoring can later yield infiltration rates that are within the correct order of magnitude. ¿ 1998 American Geophysical Union