Mid-level, mixed-phase clouds (altocumulus and altostratus) are difficult to forecast due to the fact that they are generally thin and form in areas of weak vertical velocity where operational models typically have poor vertical resolution and poor moisture initialization. This study presents experiments designed to test the utility of assimilating infrared window and water vapor channels from the Geostationary Operational Environmental Satellite (GOES) instruments, Imager and Sounder, into a mesoscale cloud-resolving model to improve model forecasts of mid-level clouds. The Regional Atmospheric Modeling Data Assimilation System (RAMDAS) is a four-dimensional variational (4-DVAR) assimilation system used to test the viability of assimilating cloudy scene radiances into a cloud-free initial model state for one case of a long-lived, isolated altocumulus cloud over the Great Plains of the United States. Observations from one observation time are assimilated and significant innovations are achieved. Three experiments are performed: (1) assimilation of the 6.7 μm (water vapor) and 10.7 μm (window) channels from GOES Imager, (2) assimilation of the 7.02 μm (water vapor) and 12.02 μm (window) channels from GOES Sounder, and (3) assimilation of the 6.7 μm channel from GOES Imager and the 7.02 μm channel from GOES Sounder. It is shown that the GOES Sounder channels provide more useful information than the GOES Imager channels due to increased sensitivity to the mid-troposphere. The decorrelation lengths and variance used in the background error covariance are varied and the impact on the results of the experiments is discussed. The effect of constraining the surface temperatures during assimilation of the window channels is also discussed. It is found that, in a cloud-free initial model state, the adjoint sensitivities are calculated on the assumption that there is no cloud, even with cloud in the satellite observations. This has a significant impact on the success of other 4-DVAR satellite data assimilation experiments.