Abstract This paper describes a new version of the University of Washington shallow cumulus parameterization. The new version includes improved treatments of lateral mixing rates into cumulus updrafts, the evaporation of precipitation and of the interaction of cumuli with the underlying subcloud layer, and a treatment of the convective inhibition-based mass-flux closure that is more numerically stable and is suitable for the long time steps of global climate models. The paper also documents its performance when combined with a new moist turbulence parameterization in simulations with version 3.5 of the Community Atmosphere Model (CAM3.5). A single-column simulation of nonprecipitating trade cumulus shows considerable improvements in vertical thermodynamic structure and less resolution sensitivity in the new schemes compared to CAM3.5. In global simulations, the new schemes, combined with an increase of vertical resolution from 26 to 30 levels, produce a significant (7%) reduction in overall climate bias, calculated from root-mean-squared error of the seasonal model climatology compared to a suite of global observations of various fields. Biases in almost all fields, particularly the shortwave cloud radiative forcing, are reduced. Geographical bias patterns in surface rainfall, liquid water path, and surface air temperature are only mildly affected by the model parameterization and vertical resolution changes.