Vertical mixing of chemical tracers and optically active constituents by deep convection affects regional and global chemical balances in the troposphere and lower stratosphere. This important process is not explicitly resolved in global and regional models and has to be parameterized. However, mixing depends strongly on the spatial structure, strength, and temporal evolution of the particular storm, complicating parameterization of this important effect in the large-scale models. To better quantify dynamic fields and associated mixing processes, we simulate a thunderstorm observed on 12 July 1996 during the STERAO-A (Stratosphere-Troposphere Experiment: Radiation, Aerosols, and Ozone) Deep Convection field project using the Goddard Cloud Ensemble (GCE) model. The 12 July STERAO-A storm had very complex temporal and spatial structure. The meteorological environment and evolution of the storm were significantly different than those of the 10 July STERAO-A storm extensively discussed in previous studies. Our 2-D and 3-D GCE model runs with uniform one-sounding initialization were unable to reproduce the full life cycle of the 12 July storm observed by the CHILL radar system. To describe the storm evolution, we modified the 3-D GCE model to include the effects of terrain and the capability of using nonuniform initial fields. We conducted a series of numerical experiments and reproduced the observed life cycle and fine spatial structure of the storm. The main characteristics of the 3-D simulation of the 12 July storm were compared with observations, with 2-D simulations of the same storm, and with the evolution of the 10 July storm. The simulated 3-D convection appears to be stronger and more realistic than in our 2-D simulations. Having developed in a less unstable environment than the 10 July 1996 STERAO-A storm, our simulation of the 12 July storm produced weaker but sustainable convection that was significantly fed by wind shear instability in the lower troposphere. The time evolution, direction, and speed of propagation of the storm were determined by interaction with the nonuniform background mesoscale flow. For example, storm intensity decreased drastically when the storm left the region with large convective available potential energy. The model appears to be successful in reproducing the rectangular four-cell structure of the convection. The distributions of convergence, vertical vorticity, and position of the inflow level in the later single-cell regime compare favorably with the airborne Doppler radar observations. This analysis allowed us to better understand the role of terrain and mesoscale circulation in the development of a midlatitude deep convective system and associated convective mixing. Wind, temperature, hydrometeor, and turbulent diffusion coefficient data from the cloud model simulations were provided for off-line3-D cloud-scale chemical transport simulations discussed in the companion paper by DeCaria et al. (2005).