Constraining intermodel spread in cloud feedback with observations is problematic because available cloud datasets are affected by spurious behavior in long-term variability. This problem is addressed by examining cloud amount in three independent ship-based [Extended Edited Cloud Reports Archive (EECRA)] and satellite-based [International Satellite Cloud Climatology Project (ISCCP) and Advanced Very High Resolution Radiometer Pathfinder Atmosphere–Extended (PATMOS-X)] observational datasets, and models from phase 5 of the Coupled Model Intercomparison Project (CMIP5). The three observational datasets show consistent cloud variability in the overlapping years of coverage (1984–2007). The long-term cloud amount change from 1954 to 2005 in ship-based observations shares many of the same features with the multimodel mean cloud amount change of 42 CMIP5 historical simulations, although the magnitude of the multimodel mean is smaller. The radiative impact of cloud changes is estimated by computing an observationally derived estimate of cloud amount feedback. The observational estimates of cloud amount feedback are statistically significant over four regions: the northeast Pacific subtropical stratocumulus region and equatorial western Pacific, where cloud amount feedback is found to be positive, and the southern central Pacific and western Indian Ocean, where cloud amount feedback is found to be negative. Multimodel mean cloud amount feedback is consistent in sign but smaller in magnitude than in observations over these four regions because models simulate weaker cloud changes. Individual models, however, can simulate cloud amount feedback of the same magnitude if not larger than observed. Focusing on the regions where models and observations agree can lead to improved understanding of the mechanisms of cloud amount changes and associated radiative impact.