Starting from multiple frequency domain measurements, this paper presents a procedure to formulate a dynamic model of a servo actuator that consists of a nominal model and an allowable model perturbation in the form of a parametric and unstructured uncertainty. A separation between parametric and unstructured uncertainty is achieved by first estimating low order linear parameter models via frequency domain curve fitting followed by a linear Principle Component Analysis (PCA) to bound the parametric variations on the estimated parameters. Remaining differences between the low order parametric models and the measured frequency responses are captured by a bounded unstructured uncertainty on a frequency dependent dual-Youla parameter that uses prior information on a stabilizing feedback controller. The resulting perturbation model is written in a standard Linear Fractional Transformation (LFT) form and the procedure is applied to experimental data obtained from several mechanically equivalent servo actuators in a Linear Tape-Open (LTO) drive.