In this article, a design procedure for robust fault isolation observers for linear systems is proposed. It allows to detect and isolate faults by means of a single, specifically parametrized observer, yielding a small implementation effort compared with standard bank-of-observers approaches. A design based on parametric eigenstructure assignment for nominal system dynamics is presented. The main contribution is to show how the remaining degrees of freedom can be utilized to optimize robustness of the fault isolation observer with respect to uncertain system parameters. Therein, the uncertainties are assumed to belong to a polyhedral domain. The application of the design to two examples verifies the results for systems defined by nominal dynamics and known bounds on the uncertain parameters.