This study evaluates the overall performances of six SCMs (CAM 3, 4, 5; GFDL AM2, 3; GISS) by comparing the simulated cloud fraction, cloud liquid and ice water content with the ARM observations at the ARM Southern Great Plains (SGP) site from January 1999 to December 2001. The evaluation focuses on the model performance under different dynamical regimes classified according to 500-hPa large-scale vertical pressure velocity. Results show that large model biases exist under strong updraft regimes: most SCMs overestimate the upper-level clouds and underestimate the lower-level clouds, which however are related to different combinations of their biases in frequency and mean cloud amount of cloudy events. Results also show distinct disagreements between models' biases in cloud fraction and cloud water content at different levels, which can be linked to deficiencies in the model parameterizations of convective and stratiform clouds. Further analyses reveal different relationships between cloud fraction and relative humidity (RH) in the models and observations. The underestimation of lower-level cloud fraction in most SCMs is mainly due to the larger threshold RH used in models, while biases in the upper-level cloud fraction exist for quite different reasons in different SCMs.