An optimal min-max instrumental variable (IV) method for estimating the parameters in difference equation models is presented. The method gives the smallest estimation error variance in the ‘worst noise’ case from a prespecified class of noises. Implementation of this new method does not require knowledge of system or noise parameters and therefore can be donewithout iteration. This is an attractive feature since other min-max or generally optimal IV methods previously introduced require at least a knowledge of system parameters, and cannot be implemented directly. Results of simulations demonstrate the applicability of the method to both artificial and real data.