An approach is presented to the problem of on-line robust control design, referred to here as adaptive calibration. It is an iterative approach which modifies the filter and model structure characteristics used in methods of system identification involving the filtered prediction error. An estimate of model uncertainty is obtained which is then used to predict closed-loop system performance with the new control if it were implemented. If predicted performance does not meet the specified performance the filters and/or model structure are modified to enhance model accuracy in the frequency range required. An analysis is presented along with an illustrative example.