During the summer convective season over the Southwest United States (commonly referred to as the monsoon), thunderstorms frequently produce damaging wind, heavy precipitation, and flash flooding. Despite the recent availability of high-resolution (~10 km grid spacing), deterministic numerical weather prediction models, prediction of these events remains difficult. This paper presents the results from a lower-resolution (~30 km grid spacing) mesoscale ensemble forecast system (EFS), primarily designed for convective quantitative precipitation forecast (QPF) guidance. During the monsoon, weak synoptic forcing and sparse observational data produce uncertainty in the initial analysis and subsequent model forecast. These uncertainties are addressed in the EFS through perturbations to the initial conditions. Model uncertainties from equally defensible model formulations are addressed through the inclusion of various convective and boundary layer parameterizations, and stochastic representations thereof. Complex terrain and mesoscale features, such as convergence from thunderstorm outflow boundaries (e.g., Wilson et al. 1992), force vertical motion at scales that the model cannot explicitly resolve. Yet grid resolvable vertical velocity is the main component of the cumulus parameterization schemes (e.g. Kain and Fritsch 1993). In our EFS, this subgrid scale forcing is represented by stochastic perturbations with a convective time scale. Results for QPF from various ensemble configurations are presented, and their impact on ensemble dispersion and predictability limits are discussed.