An approach is presented for modeling uncertainty from a priori information of a plant. A feasible model set is defined of all systems governed by a class of differential equations with certain norm bounds of unknown input and output operators and partially known eigenparameters. Then an unstructured uncertainty class of models including the feasible model set is characterized with least magnitude of additive perturbation in frequency domain. An example of finite dimensional compensator design is demonstrated to illustrate the usefulness of the approach.