The objective of this paper is to investigate the backscattering from multiscale soil surfaces through a comparison of analytical model predictions with numerical computations and experimental data. A numerical simulation of the backscatter from multiscale surfaces is performed in order to assess the sensitivity of the radar backscatter to the multiscale effect as a function of the roughness range of the surfaces, the polarization, and the incidence angle of the impinging electromagnetic wave. The numerical backscattering coefficients are compared with those obtained using the traditional integral equation method (IEM) as well as those obtained using a novel version of the IEM which has been adapted to model the backscattering from multiscale surfaces (MLS IEM). As a result, the surface conditions under which the traditional IEM can still predict well the backscatter from multiscale surfaces as well as the conditions under which traditional IEM fails are derived. Subsequently, 25 m long roughness profiles collected over the Marestaing test site located near Toulouse, France, are analyzed in order to estimate input parameters for the two versions of the IEM model. Then, a comparison is made between the single-scale IEM and MLS IEM predictions and ERS synthetic aperture radar (SAR) data acquired over the site. The results show that over medium-rough soils (i.e., harrowed soils) the SLS IEM is in good agreement (within 1 dB) with ERS SAR data, whereas MLS IEM underestimates the measurements of 2--3 dB. In contrast, over very smooth soils (i.e., rolled soils) the MLS IEM is in overall better agreement with ERS SAR data than SLS IEM. The latter overestimates the measurements of 3 dB. ¿ 2001 American Geophysical Union