This paper presents a comprehensive study of the near-bed hydrodynamics at non-moving streambeds based on laboratory experiments in open-channel flows. Pressure and velocity measurements were made with an array of up to 15 miniaturized piezo-resistive pressure sensors within the bed and slightly above it, and a two-dimensional particle-image-velocimetry (PIV) system measuring in streamwise vertical or horizontal planes. Three different types of bed materials were studied covering typical natural streambed conditions. The range of the global Reynolds number covered in the experiments was from 20000 to 200000. This study provides new insights into the flow structure over gravel beds based on the PIV measurements in both streamwise vertical and horizontal planes. In a streamwise vertical plane, large-scale wedge-like flow structures were observed where a zone of faster fluid over-rolled a zone with slower fluid. The resulting shear layer was inclined along the flow at an angle of 10°–25° to the bed, and was populated with clockwise rotating eddies. This mechanism occurred with sufficient frequency and shape to leave an ‘imprint’ in the velocity statistics. Typically, the described flow pattern is formed near the bed and is approximately scaled with the height of the logarithmic layer, although the biggest structures extended over the whole flow depth. In a horizontal near-bed plane, turbulent structures formed a patched ‘chessboard’ pattern with regions of lower and higher velocities that were elongated in the streamwise direction. Their lateral extension was typically two to four times the equivalent sand roughness with lengths up to several water depths. The dimensions of the regions were increasing linearly with the distance from the bed. These findings are consistent with conceptual models originally developed for smooth-wall flows. They also support observations made in rough-bed flume experiments, numerical simulations and natural rivers. Spatial fields of bed-pressure fluctuations were reconstructed by applying Taylor's frozen turbulence hypothesis on time data obtained with an array of pressure sensors. Based on the conditional sampling of velocity patterns associated with pressure-drop events a distinct bed-destabilizing flow-pressure pattern was identified. If a high-speed fluid in the wake of a large-scale wedge-like flow structure reaches the vicinity of the bed, a phenomenon akin to a Bernoulli effect leads to a distinctive low-pressure pattern. The resulting force may exceed the particles' submerged weight and is assumed to be able to give an initial lift to the particle. As a result, the exposed area of a particle is amplified and its angle of repose is reduced, increasing the probability for entrainment.