The performance of several schemes for diagnosing cloud cover from forecast model output was tested using a global numerical weather prediction model and the operational USAF RTNEPH (real-time nephanalysis) cloud analysis. In the present study, schemes were developed from cloud cover statistics stratified by synoptic weather regime. The synoptic regime were defined in terms of vertical profiles of temperature, winds, and moisture. The meteorological significance of these regimes was illustrated by relating them to synoptic features. The simplest scheme (AVG) assigned the average cloud cover to each of the regimes; a variant of the cloud curve algorithm (CCA) technique was developed in which separate cloud-RH curves were derived for each regime by a mapping of the cumulative frequency distribution of RH and cloud cover. Their performance was compared against a number of other diagnostic schemes, including a multiple linear regression method that used global regression equations for cloud cover from a large number of atmospheric and geographic predictors; a version of the Slingo scheme; and simple persistence. Results indicate that the schemes with the lowest rms errors (AVG, and the regression scheme) also had highly unrealistic frequency distributions, with too few points that were close to either clear or overcast values. Persistence was found to provide competitive or superior forecasts out to 24–36 h. The applicability of these results to improved model and cloud observations is discussed.