Soil moisture is a key component in landsurface parameterization schemes as it is closely related to evaporation and plays an important role in the hydrological cycle and in soil-vegetation biochemistry and determines the partition of precipitation into runoff and evaporation and the net available energy into latent and sensible heat fluxes. One of the major research activities in the PILPS/AMIP subproject (No. 12) is to understand current landsurface parameterizations used in climate models and to assess their performances. As part of this evaluation 14 landsuface schemes conducted 15 numerical experiments using the atmospheric forcing and surface parameters derived from HAPEX-MOBILHY and the performance of the schemes were evaluated against these observational data. The intercomparison of the simulations revealed a large difference of about 200 mm in total soil moisture content for a 1.6 m soil layer. After careful adjustment of the parameters characterizing the soil hydraulic properties and those of surface properties, the disagreement between the models decreased to 70 mm for the bare soil period and about 100 mm in the growing season, which is still quite large. All models could correctly describe the annual trend of soil moisture in a qualitative way, but compared to observation most schemes underpredicted the total soil water content, especially during the growing season. In contrast to the total soil layer, models generally overestimated the soil water content in the root zone. Most schemes conserve water well, but the soil moisture budget is achieved in very different ways in different models. The ranges of annual evaporation as well as runoff-drainage are as large as 250 mm. The partitioning of surface energy into latent and sensible fluxes is closely related to the partitioning of precipitation into evaporation and runoff-drainage. Current evaluations of soil moisture are unreliable even with carefully specified landsurface parameters and accurate atmospheric forcing. The prediction of soil moisture in global models seems to be extremely uncertain at present.