The sensitivity of all-sky infrared radiances to both hydrometeor content and cloud height leads to a very non-Gaussian distribution of first-guess departures. This non-Gaussianity can be mitigated by the application of cloud-dependent error models that normalize departures by an estimate of the cloud-height effect via assigning increased errors in situations with high clouds that can lead to very large departures. In the current study, we systematically evaluate existing error models and propose a revised approach that leads to a better fit to a Gaussian distribution at no additional cost. Furthermore, the revised approach is physically better justified as the cloud effect is estimated by the maximum cloud effect of model and observations, which determines the largest possible departure. Preceding studies, in contrast, used the mean cloud effect of model and observations.</p> <p>Our study is based on a one-month data set of infrared observations in two water vapor channels (6.2 and 7.3 μm) from the Spinning Enhanced Visible and Infrared Imager (SEVIRI) onboard the Meteosat Second Generation satellite and corresponding simulations from the weather forecast model AROME (Application of Research to Operations at Mesoscale) over central Europe. The evaluation of different approaches revealed that near-Gaussian departures can be achieved with three different approaches for the estimation of the cloud effect: (1) deviation from a climatologically estimated value; (2) deviation from the clear-sky brightness temperature of a window channel; and (3) deviation from the clear-sky brightness temperature of the channel that is assimilated. For the lower-peaking channel (7.3 μm) with a larger cloud effect, best results were achieved with the first two options. For the 6.2 μm channel, the third option led to a slightly more Gaussian distribution. The third option, however, requires a quality control that eliminates about 10 % of the observations.