During a manoeuvre it is frequently found that a ship's yaw dynamics vary widely. To obtain a consistent response to the helm, it is therefore advisable to provide an autopilot to cope with plant uncertainties. Using the method of Horowitz, a high-loop-gain linear system may be designed which achieves an acceptable spread of transient responses. Unfortunately, there is a penalty to be paid in the form of excessive rudder activity when course keeping. It is therefore desirable to find a way of reducing the loop-gain requirement by attempting to remove some of the plant ignorance. The paper therefore proposes to reduce the plant uncertainty by using an inverse nonlinear compensator derived from a simple linear time-invariant model of the ship. It is then possible to control the resultant modified (and of reduced ignorance) plant with a much smaller loop gain. The paper shows that the design results in consistent manoeuvring performance coupled with much reduced rudder activity during course keeping. The complete compensator is readily realisable in DDC.