A new method is described for introducing integral action into linear quadratic optimal control systems. The method is particularly appropriate for thes-domain design of controllers for output-feedback optimal systems. The usual performance criterion is modified by the addition of an integral operator and this ensures that the system has zero steady-state error. The technique has several advantages over those previously discussed, and in its simplest form may be shown to be equivalent to the use of a [udot] term in tho cost function. Optimal output-feedback controllers are defined for both stochastic and deterministic multivariable systems.