For at least 20 years, nadir altimetry satellite missions have been successfully used to first monitor the surface elevation of oceans and, shortly after, of large rivers and lakes. For the last 5-10 years, few studies have demonstrated the possibility to also observe smaller water bodies than previously thought feasible (river smaller than 500 m wide and lake below 10 km 2). The present study aims at quantifying the nadir altimetry performance over a medium river (200 m or lower wide) with a pluvio-nival regime in a temperate climate (the Garonne River, France). Three altimetry missions have been considered: ENVISAT (from 2002 to 2010), Jason-2 (from 2008 to 2014) and SARAL (from 2013 to 2014). Compared to nearby in situ gages, ENVISAT and Jason-2 observations over the lower Garonne River mainstream (110 km upstream of the estuary) have the smallest errors, with water elevation anomalies root mean square errors (RMSE) around 50 cm and 20 cm, respectively. The few ENVISAT upstream measurements have RMSE ranging from 80 cm to 160 cm. Over the estuary, ENVISAT and SARAL water elevation anomalies RMSE are around 30 cm and 10 cm, respectively. The most recent altimetry mission, SARAL, does not provide river elevation measurements for most satellite overflights of the river mainstream. The altimeter remains "locked" on the top of surrounding hilly areas and does not observe the steep-sided river valley, which could be 50 m to 100 m lower. This phenomenon is also observed, for fewer dates, on Jason-2 and ENVISAT measurements. In these cases, the measurement is not "erroneous", it just does not correspond to water elevation of the river that is covered by the satellite. ENVISAT is less prone to get "locked" on the top of the topography due to some differences in the instrument measurement parameters, trading lower accuracy for more useful measurements. Such problems are specific to continental surfaces (or near the coasts), but are not observed over the open oceans, which are flatter. To overcome this issue, an experimental instrument operating mode, called the DIODE/DEM tracking mode, has been developed by CNES (Centre National d"Etudes Spatiales) and has been tested during few Jason-2 cycles and during the first SARAL/AltiKA cycle. This 3 tracking mode "forces" the instrument to observe a target of interest, i.e. water bodies. The example of the Garonne River shows, for one SARAL ground track, the benefit of the DIODE/DEM tracking mode for a steep-sided river reach, which is not detected using the nominal instrument operating mode. Yet, this mode relies on ancillary datasets (a priori global DEM and global land/water mask), which are critical to obtain river valley observation. The ultimately computed elevations along the satellite tracks, loaded on board, should have an absolute vertical accuracy around 10 m (or better). This case also shows, when the instrument is correctly observing the river valley, that the altimeter can detect water bodies narrower than 100 m (like an artificial canal). In agreement with recent ...