Discharge measurement is a critical task for gravel-bed channels. Under high-flow conditions, the elevation of the riverbed changes significantly by intensive torrential flow. The stage-discharge relations commonly used for stream discharge estimation may no longer be adequate. The contact-type velocity measuring is also subject to measurement errors and/or instrument failures by the high-flow velocities, driftwood, stumps, and debris. This study developed a new real-time method to estimate river discharge in gravel-bed channels. A systematic measuring technology combining ground-penetrating radar and surface-velocity radar was employed. The rating curves representing the relations of water surface velocity to the channel cross-sectional mean velocity and flow area were established. Stream discharge was then deduced from the resulting mean velocity and flow area. The proposed method was examined in a steep gravel-bed reach of the Cho-Shui River in central Taiwan. The estimated stream discharge during three flood events were compared to the prediction by using the stage-discharge relation and the index-velocity method.The proposed method of this study is capable of computing reasonable values of discharge for an entire flood hydrograph, whereas the other two methods tend to produce large extrapolation errors. Moreover, when the computed discharge is used in 2D flood flow simulation, the proposed method demonstrates better performance than the commonly used stage-discharge and index-velocity methods.