Using multispectral SPOT images and 1/100,000 topographic data, we present an improved map of the active Red River fault zone between Midu (Yunnan, China) and Hanoi (Vietnam). The fault zone is composed of parallel strands, one of which, the Yuanjiang fault was previously undetected. There also appears to be a component of extension all along the fault zone. Such extension increases toward the SE, from Yunnan to the south China sea coast, and the vector describing the motion of south China relative to Indochina points within the N45¿--135 ¿E quadrant. We attempt to assess the Plio-Quaternary dextral slip rate on the Red River fault (RRF) by restoring large river offsets and searching for the largest, plausible one. Across much of Yunnan, the fault is perpendicular to local catchments that drain into the Red River. From precise mapping of the river courses on SPOT satellite images and on 1/100,000 topographic maps, numerous multiple offsets along the fault can be detected and reconstructed. The lack of correlation between the apparent offsets and the lengths of the rivers upstream from the fault suggests either that the drainage system was in large part established prior to the onset of dextral slip along the fault or that frequent captures have occurred. We thus try to find the best fit between series of river channels upstream and downstream from the fault by progressively restoring the dextral displacement in increments of 500 m, up to an offset of 50 km. For each increment we measure the misfits (root mean squares, RMS) between the upstream and downstream channels. The best fit and smallest RMS are obtained for an offset of 25¿0.5 km that we interpret to represent the clearest, large right-lateral displacement recorded in the geomorphology along the active Red River fault. Since dextral motion is likely to have started around 5 Myr, the most probable average Plio-Quaternary slip rate on the fault is of order of 5 mm/yr. We attribute the apparent lack of seismic activity on a large stretch of the fault to millennial recurrence times between great earthquakes. Our study shows that relatively small drainage systems can keep a good record of fairly large cumulative fault offsets. ¿ 2001 American Geophysical Union