This work explores the feasibility of variational assimilation of radar data to improve the forecast of a mesoscale model. The model is heritage of the CSU Regional Atmospheric Modeling System (RAMS) and has full coupling between all relevant atmospheric processes- dynamical, microphysical and radiative. The model is used to simulate an ice cloud, and results are found to be sensitive to the choice of large-scale forcing. The adjont of the Cloud Resolving Model is used to understand model response to variation in specified inputs, such as initial and boundary conditions. Synthetic radar reflectivity observations are used to test the feasibility of variational data assimilation using boundary conditions (BC) as control variable. Radar reflectivity is computed from the simulated cloud fields introducing a radar observational operator. The model is then run with perturbed BCs, producing a cloud field different from the original. A four--dimensional variational system is used to recover ``true'' BCs. Experiments show that the assimilation system is successful in this attempt. A consequence of these experiments is that assimilation of cloud radar data is feasible and has, in principle, the potential to ``correct'' the model forecast.