Results show that while the CMIP5 models simulate the tropical TOA CREs reasonably well, they are rather problematic in representing tropical clouds. The good simulations of the TOA CREs are often a result of compensating errors between different cloud processes. Climatologically, the CMIP5 models produce considerably less yet optically thicker clouds than the observed. The models particularly show strong underestimation of mid- and low-level clouds. While the tropical mean high-level clouds in the majority of the models are moderately less than that of the observed, the models consistently produce less high clouds over the Maritime continent and more high clouds over the western tropical Indian Ocean and subtropical Pacific trade cumulus regions. The reasonably good model simulations of the tropical mean TOA CREs are essentially a result of compensating errors over different tropical dynamical regimes and different cloud regimes. The model biases in climatology strongly affect the model simulations of climate variability. Taking ENSO as an example, the CMIP5 models consistently underestimate the TOA CRE anomalies over the Maritime Continent due to the notable model underestimation of high clouds there. In the central tropical Pacific, while the CMIP5 models simulate the TOA shortwave and longwave CRE anomalies fairly well, they are unfortunately a result of compensating errors between an underestimation of CRE changes from thin high clouds and an overestimation of CRE changes from medium and thick high clouds.