Information on cloud properties from ATSR-2 radiances originates from the interaction of solar and terrestrial radiation with the cloud. Due to the spectral dependence of cloud optical properties, observations in different channels of ATSR-2 can be combined to provide information on a cloud phase, top height, effective radius cloud fraction and optical depth. These parameters are retrieved using an optimal estimation approach which enables us to extract information from all channels simultaneously. The method allows us to characterize the error on each parameter assuming the observed cloud is consistent with the modelled cloud (i.e. reasonably plane-parallel in nature). It is also possible to identify when model assumptions are inappropriate by examining the fit between simulated and observed radiances. This technique has been used in the NERC GRAPE (Global Retrieval of Clouds and Evaluation) project. In this paper the optimal estimation algorithm will be outlined and products and validation from the 5 year data set will be presented.