Land surface scheme simulations, which differ only in their parameterization of surface runoff, are performed for the Amazon and Mississippi River basins at ~1.88¿¿1.90¿ resolution. In the first simulation, soil moisture and precipitation intensity are assumed to be uniformly distributed. In the second simulation, precipitation intensity is assumed to be exponentially distributed, and soil moisture subgrid variability is expressed in terms of the variable water-holding capacity of the grid cell. The effect of subgrid variability of precipitation intensity and soil moisture is taken into account to model surface runoff more realistically. National Center for Environmental Prediction (NCEP) reanalysis data with its precipitation estimates adjusted to long-term values are used to drive the land surface scheme, and flow routing is performed to obtain streamflow at three locations in each river basin. Comparisons with observations indicate that the inclusion of subgrid variability results in slightly improved streamflow simulations in terms of reduced root-mean-square errors and better correlation coefficients. Sub-grid-scale variability also results in significant changes in the magnitude, time, and frequency of surface runoff generation, partitioning of total runoff into surface runoff and drainage and slightly drier soil moisture conditions. ¿ 2001 American Geophysical Union