The global statistics of the diurnal cycle of warm-season precipitation simulated by NASA's Seasonal to Interannual Prediction Project (NASA/NSIPP) atmospheric general circulation model (AGCM) were evaluated using the Tropical Rainfall Measuring Mission (TRMM) Microwave Imager (TMI) rain retrievals. The model has notable biases in the phase of the diurnal cycle over land, where it produces rainfall maxima in the early afternoon, several hours earlier than the observed evening maxima. The model also produces too little precipitation in the nighttime over land. Similar biases in the phase were found over ocean, although the statistics are less robust. An analysis of the convective and stratiform contributions to the precipitation indicates that the incorrect representation of the diurnal cycle is primarily tied to deficiencies in the deep convection scheme. A set of sensitivity experiments shows that the phase of the maximum diurnal precipitation is quite sensitive to the change of the convection starting (parcel origination) level and the increase of the convection adjustment (relaxation) timescale in the Relaxed Arakawa-Schubert scheme. Both modifications act to delay the timing for the maximum development of CAPE, which lead to improvements in the diurnal cycle of precipitation, especially in correcting the phase errors over land, although the modifications tend to reduce the diurnal amplitudes substantially. The study suggests that improvements to the diurnal cycle in current models require improvements to the parameterized deep convection schemes, including the coupling with the boundary layer, the characteristic timescale of convection adjustment, and the triggering process for nocturnal precipitation.