The effect of surface sensible and latent heat flux heterogeneity on the daytime changes in the potential for development of deep convection is evaluated by a combined data analyses and numerical modeling approach. The study domain is the CART (Cloud and Radiation Testbed) site of the Atmospheric Radiation Measurement Program in the Southern Great Plains and the time period of the study is the summer of 1995. The 300 km x 350 km study domain is characterized by substantially different surface characteristics due, in part, to the harvest of extensive areas of winter wheat and to spatial variations in rainfall over the region. Meteorological variables obtained form a dense measurement network in combination with satellite information are used to drive the Sib2 land surface parameterization scheme to generate spatially and temporally varying surface sensible and latent heat fluxes. The flux values are checked against measured heat fluxes from a suit of Bowen ratio stations distributed across the site and good agreement are found between the Sib2 calculations and the observations. A high resolution mesoscale model is modified so that it can be driven by prescribed surface heat flux values and simulations are carried out with the model driven by both variable and uniform surface flux distributions across the study domain for selected cases when the large-scale environment was relatively unstable. Several indicators of deep convection, including the Lifted Index, the Convective Temperature, and the Convective Available Potential Energy (CAPE), are calculated for the entire CART domain using the modeled soundings and their sensitivity to surface flux variations are examined. The implications of these sub-grid scale boundary-layer effects for climate model simulations of convective precipitation are discussed.