Conventional measurements of river flows are costly, time‐consuming, and frequently dangerous. This report evaluates the use of a continuous wave microwave radar, a monostatic UHF Doppler radar, a pulsed Doppler microwave radar, and a ground‐penetrating radar to measure river flows continuously over long periods and without touching the water with any instruments. The experiments duplicate the flow records from conventional stream gauging stations on the San Joaquin River in California and the Cowlitz River in Washington. The purpose of the experiments was to directly measure the parameters necessary to compute flow: surface velocity (converted to mean velocity) and cross‐sectional area, thereby avoiding the uncertainty, complexity, and cost of maintaining rating curves. River channel cross sections were measured by ground‐penetrating radar suspended above the river. River surface water velocity was obtained by Bragg scattering of microwave and UHF Doppler radars, and the surface velocity data were converted to mean velocity on the basis of detailed velocity profiles measured by current meters and hydroacoustic instruments. Experiments using these radars to acquire a continuous record of flow were conducted for 4 weeks on the San Joaquin River and for 16 weeks on the Cowlitz River. At the San Joaquin River the radar noncontact measurements produced discharges more than 20% higher than the other independent measurements in the early part of the experiment. After the first 3 days, the noncontact radar discharge measurements were within 5% of the rating values. On the Cowlitz River at Castle Rock, correlation coefficients between the USGS stream gauging station rating curve discharge and discharge computed from three different Doppler radar systems and GPR data over the 16 week experiment were 0.883, 0.969, and 0.992. Noncontact radar results were within a few percent of discharge values obtained by gauging station, current meter, and hydroacoustic methods. Time series of surface velocity obtained by different radars in the Cowlitz River experiment also show small‐amplitude pulsations not found in stage records that reflect tidal energy at the gauging station. Noncontact discharge measurements made during a flood on 30 January 2004 agreed with the rated discharge to within 5%. Measurement at both field sites confirm that lognormal velocity profiles exist for a wide range of flows in these rivers, and mean velocity is approximately 0.85 times measured surface velocity. Noncontact methods of flow measurement appear to (1) be as accurate as conventional methods, (2) obtain data when standard contact methods are dangerous or cannot be obtained, and (3) provide insight into flow dynamics not available from detailed stage records alone.