Most NWP models now have one or two prognostic variables such as cloud fraction and cloud water content to represent clouds. However, although total cloud cover can be derived from satellite, there are few observations available to evaluate the vertical profiles of cloud cover and water content held in these models. In the CLOUDNET project (BAMS article, June 2007) vertically pointing radar and lidars have been operated continuously at three sites in Europe for several years, providing profiles with 30second/60m resolution, and the cloud properties inferred from these observations have been compared with the vertical profile of the cloud variables held in seven different operational models each hour for the grid boxes over each site. Algorithms have been developed which categorise the targets seen by the active sensors as cloud/no-cloud, and when cloud if it is liquid, ice or mixed phase. Once each 60m/30second pixel has been categorised, then the cloud fraction is the number of such pixels in the hourly model grid box which have been deemed to be cloudy. Radar signals are corrected for gaseous and liquid attenuation and then converted into an ice water content. Liquid water path, derived from radiometers, is converted to liquid water content using the cloud base as detected by lidar and the cloud-top given by the radar. The comparisons demonstrate that the NWP model performance is generally good, but we can identify particular biases and errors in each model. The same analysis is being applied to the ARM sites and results from this study will also be presented.