A representation for deterministic disturbances which is consistent with the usual representation for stochastic disturbances is given. Using the internal-model principle, a rearrangement of the estimation model which prefilters the input data is proposed. For the case of unknown deterministic disturbances, the prefilter parameters are adjusted adaptively. The proposed estimation model results in a parameter vector that does not contain the critical poles of the controller. Hence the risks of the controller having poles in the unstable region during the transient period are greatly reduced.