We analyze the causes of the large thermal biases in the simulation of the National Center for Atmospheric Research's Community Climate Model Version 3 (CCM3) single-column model (SCM) when it is forced with the Atmospheric Radiation Measurement Program's (ARM) summer 1995 Intensive Observing Period (IOP) data. We have found that deficiencies of the convection triggering function used in the model can explain a significant part of the biases. In the model, convection is triggered whenever there is positive convective available potential energy (CAPE), which happens to occur during daytime due to solar heating of the land surface. In observations, however, convection takes place only when the large-scale dynamic condition is favorable to the release of CAPE. On the basis of observations we propose a new triggering function for the CCM3 deep convection scheme. We assume that model convection occurs only when the large-scale dynamic forcing makes positive contributions to the existing positive CAPE. Improved simulation results are obtained when the new triggering function is implemented in the model. We further test the new triggering function using the ARM Southern Great Plains summer 1997 IOP data and the Global Atmospheric Research Program's Atlantic Tropical Experiment phase III observations. These tests confirm the results obtained from the ARM 1995 experiments. ¿ 2000 American Geophysical Union