In motion or process control systems, a variety of design techniques have been proposed because of the demand for high performance. The higher performance we demand, the higher the degree of the controller becomes. The controller is generally designed by a CAD system and implemented with a microprocessor. But the microprocessor does not have enough precision to realize the results of design by the CAD system. Therefore, the system performance is degraded by finite word length (FWL) effects. To deal with FWL problems, many design methods have been considered in the signal processing field, and high-ordered digital filters are often used. Among these methods, the implementation technique based on the state-space realization can minimize the sensitivity to perturbation of coefficients. Noting that optimal realizations with the same transfer function are unique only up to an orthogonal similarity transformation, we must choose the realization within this class of optimal realizations. In this paper, we present an algorithm to find a state-space realization which minimizes the frequency-weighted sensitivity measure of the controller performance. Furthermore, we present some experimental results to verify the effectiveness of the proposed algorithm.