Guided by ground-based radar and lidar profiling at the Barbados Cloud Observatory (BCO), this study evaluates trade-wind cloudiness in ECMWF's Integrated Forecast System (IFS) and nine CMIP5 models using their single-timestep output at selected grid points. The observed profile of cloudiness is relatively evenly distributed between two important height levels: the lifting condensation level (LCL) and the tops of the deepest cumuli near the trade-wind inversion (2–3 km). Cloudiness at the LCL dominates the total cloud cover because it is nearly always present. Variance in cloudiness is larger near the inversion. The IFS reproduces the height of the clouds and the variability, but underestimates cloudiness at both the LCL and the inversion. A few CMIP5 models produce a single stratocumulus-like layer near the LCL, but more than half of the CMIP5 models reproduce the observed cloud layer depth in long-term mean profiles. At individual time steps, however, half of the models either form cloud near the climatological inversion or near the LCL instead of at both. In seven models, no clouds are produced at either level 10 to 65% of the time, compared to 3% in the observations. Models therefore tend to overestimate variance in cloudiness near the LCL. This variance is associated with longer time scales than in observations, which suggests that modeled cloudiness is too sensitive to large-scale processes. To conclude, many models do not appear to capture the processes that underlie changes in cloudiness, which is relevant for cloud feedbacks and climate prediction.