Soil erosion may be apprehended at different scales and different time intervals in order to quantify the transport of suspended sediment at the outlet of large watersheds. In the Walloon region, several projects were conducted to estimate the erosion rates in watersheds ranging from 16 to 2900 km². Both discrete sampling methods (allowing the coverage of a large area through field campaigns during major floods or moderate hydrological events) and continuous turbidity measurements devices (studying a small number of locations with a small sampling interval and a wide range of sampled flow rates) were tested in these projects. At the region scale (Figure 1), the mean annual sediment erosion rate reaches several hundreds of tons per square kilometre and per year in the loess belt (Senne, Dyle and Gette watersheds) with a huge sensibility to extreme hydrological events while the mean annual sediment transport value reaches 20 t.km-2.yr-1 in Lorraine, 34 t.km-2.yr-1 in Ardenne and 69 t.km-2.yr-1 in Entre-Vesdre-et-Meuse. The aims of our researches are 1) the determination of the effect of the sampling frequency on the annual sediment transport rate estimation ; 2) the definition of the prerequisites to perform efficient turbidity measurements in rivers with high suspended load concentrations ; 3) the identification of the sources of errors due to the interpolation methods through a comparison of several methods proposed in the literature allowing the quantification of long-term erosion rates using a small number of discrete samples ; 4) the observation and explanation of clockwise and anti-clockwise hysteresis loops in relationship with the sequence of floods occurring in large watersheds and 5) the highlighting of the regional differentiation in the erosion rates, particle size and proportion of organic matter in the suspended load samples due to spatial variations of the soil substrate and the land cover characteristics.