The sensitivity of the hydrological cycle to soil hydrology is investigated with the LMD GCM. The reference simulation includes the land-surface scheme SECHIBA, with a two-reservoir scheme for soil water storage and runoff at saturation. We studied a non-linear drainage parametrization, and a distributed surface runoff parametrization, accounting for the subgrid scale variability (SSV) of soil moisture capacity, through a distribution where the shape parameter was b. GCM results show that the drainage parametrization induces significant reductions in soil moisture and evaporation rate compared to the reference simulation. They are related to changes in moisture convergence in the tropics, and to a precipitation decrease in the extratropics. When drainage is implemented, the effect of the SSV parametrization (b=0.2) is also to reduce soil moisture and evaporation rates compared to the simulation with drainage only. These changes are much smaller than the former, but the sensitivity of the hydrological cycle to the SSV parametrization is shown to be larger in dry periods, and to be enhanced by an increase of the shape parameter b. The comparison of simulated total runoffs with observed data shows that the soil hydrological parametrizations does not reduce the GCM systematic errors in the annual water balance, but that they can improve the representation of the total runoff's annual cycle.