A state-of-the-art mesoscale atmospheric model was used to investigate the three-dimensional structure and evolution of shallow convective clouds and precipitation in heterogeneous and homogeneous domains. In general, the spatial distribution of clouds and precipitation is strongly affected by the landscape structure. When the domain is homogeneous, they appear to be randomly distributed. However, when the landscape structure triggers the formation of mesoscale circulations, they concentrate in the originally dry part of the domain, creating a negative feedback which tends to eliminate the landscape discontinuities, and spatially homogenize the land water content. The land surface wetness heterogeneity of the domain and the total amount of water vapor present in the atmosphere (locally evapotranspired and/or advected) affect the precipitation regime. In general, the upward motion of mesoscale circulations generated by landscape heterogeneities is stronger than thermal cells induced by turbulence. Furthermore, their ability to transport moist, warm air to higher elevations increases the amount of water that can be condensed and precipitated. The evolution of shallow convective clouds and precipitation consists of a ''build-up phase'' during which turbulence is predominant and responsible for the moistening of the atmosphere. In heterogeneous domains, it is also responsible for the creation of horizontal pressure gradients leading to the generation of mesoscale circulations. This phase occurs during the morning hours. From about 1200 until 1600 LST, clouds develop and most of the precipitation is produced. This is the ''active phase.'' After 1600 LST, the horizontal thermal and pressure gradients, which fed the energy necessary to create and sustain the mesoscale circulations, gradually disappear. This is the ''dissipation phase.'' The differences and similarities obtained between three-dimensional and two-dimensional simulations were also studied. These simulations indicate that, unless the landscape presents a clear two-dimensional structure, the use of such a two-dimensional model is not appropriate to simulate this type of clouds and precipitation. Conversely, two-dimensional simulations can be confidently used, provided that the simulated domain presents a two-dimensional heterogeneity. ¿ American Geophysical Union 1996