A new method to derive a cloud type climatology is applied to cloud observations over the Southern Great Plains (SGP) Atmospheric Radiation Measurement (ARM) site and to ECMWF model forecasts, in order to compare model and radar derived cloud type statistics and identify the major deficiencies in model cloud vertical distribution. The results indicate that cirrus and to a lesser extent middle level clouds are the major cloud types missing in the model simulations and that they are missing mostly as parts of multi-type rather than single-type structures. Boundary layer clouds are simulated at approximately the right amounts in the annual mean statistics, but this result comes from the model simulating too little boundary layer cloud in the winter and too much in the summer. Overall, the model forecasts miss about 11% of the total cloud amount, and most of the missing cloud occurs at time periods when multiple cloud types are present in the observations.