Geophysical models are more and more complex and can describe, in a very careful way, the geodynamical processes. These models depend on several geometrical and physical parameters, which characterize the models behaviour. These parameters are not errorless since they are known with some uncertainties. Thus, model predictions are affected by these parameter uncertainties. Usually, the model errors are not taken into account and are not propagated to the estimated quantities. This can cause improper hypothesis testing when geodetic data and model predictions are compared. In this paper, a method is presented which allows considering the model errors through the definition of the spatial covariance function of the model. A simulation is set up to prove the method feasibility.