A simple method of optimally reducing the order of systems with time delays is proposed. The performance indices used for optimization are the integrals of the time-weighted squared error between the responses of the reduced-order and original models. The performance indices are first expressed in terms of the reduced order system unknown parameters and a minimization of these indices gives the reduced model optimal parameters. Cheap and accurate computation of optimal reduced order models with time delay has so far proved abortive as closed-form expressions for these indices have proved difficult to obtain. Simple feedback controllers parametrized by the internal model control method using the reduced models result in excellent performance when implemented on the original model.