A new method for the design of a linear-quadratic optimal regulator directly from input-output data is proposed. The method can be used to design a multivariable regulator when the plant is open-loop stable, with more outputs than inputs, and is left-invertible. The major advantages of the method are threefold. First, it involves an output feedback control law; hence, no state estimation is needed for implementation. Second, the computation of this controller is possible without explicit identification of the plant model parameter. Third, a reduced-order controller, if one exists, may be synthesized without additional effort to reduce the plant model.