Water surface roughness strongly impacts microwave backscattering process over rivers. Therefore, developping radar interferometry techniques over continental waters to determine river slope (cross-track interferometry) or surface velocity (along-track interferometry) requires a detailed characterization of water surface roughness, its relation with river flow and wind conditions, its influence on backscatter coefficient. In situ measurement of river surface roughness is a complex task as surface topography is rapidly changing and sensitive to obstacles or contact measurements. Laboratory measurements, although highly informative, fall short to represent the diversity of river flow and wind conditions. In the framework of the SWOT mission preparatory phase, a method was developped for field measurement of river surface roughness. It was tested and validated in laboratory conditions, and implemented in natural conditions on the Rhône river, under various wind intensities. The method is based on the acquisition of water pressure time series by a network of immersed pressure sensors, with synchronized 10 Hz sampling and 0,1 mbar accuracy (1mm water elevation). A preliminary low frequency filtering is applied to each sensor pressure time series to remove water level trends. Mean depth h, frequency spectrum, dominant frequency f and standard deviation of water pressure σP are determined per time interval (180s). A multisensor analysis is realised to determine directional wave celerity c and directional correlation length Lcorr. Finally standard deviation of surface water level σZ is determined for each sensor by applying a correction factor, taking into account the signal damping with depth. σZ=e^(2.pi.h.f/c).σP(h). Laboratory tests validated the measurement technique, provided that the correction factor does not exceed 10 (i.e. the immersed sensor records more than 10% of the surface signal) which is achievable by limiting the sensor depth. Field measurements were realised during a campaign on the Rhône river (may 2011). The sensor network was located 10m from the river bank and 0,15m under the surface water. It recorded during 7 hours at 10 Hz. Wind measured at 2m height had the same direction as river flow (southward), with intensity changing progressively from 8m/s to 0m/s. Surface roughness was characterized, per 3 minute time intervals, by Frequency spectrum, dominant frequency, standard deviation of surface water level σZ, directional wave celerity c and directional correlation distance Lcorr. Its sensitivity to wind conditions was analysed and modelled. The method is currently being adapted on a floating support for intensive river surface roughness measurement.