A variational method to retrieve profiles of temperature, humidity, and cloud is described, which combines observations from a 12-channel microwave radiometer, an infrared radiometer, and surface sensors with background from shortrange numerical weather prediction (NWP) forecasts in an optimal way, accounting for their error characteristics. An analysis is presented of the error budget of the background and observations, including radiometric, modeling, and representativeness errors. Observation errors of some moisture channels are found to be dominated by representativeness, due to their sensitivity to atmospheric variability on smaller scales than the NWP model grid, whereas channels providing information on temperature in the lowest 1 km are dominated by instrument noise. Profiles of temperature and a novel total water control variable are retrieved from synthetic data using Newtonian iteration. An error analysis shows that these are expected to improve mesoscale NWP, retrieving temperature and humidity profiles up to 4 km with uncertainties of 1 K and 40% and 2.8 and 1.8 degrees of freedom for signal, respectively, albeit with poor vertical resolution. A cloud classification scheme is introduced to address convergence problems and better constrain the retrievals. This Bayesian retrieval method can be extended to incorporate observations from other instruments to form a basis for future integrated profiling systems.