Automatic guidance of farm tractors would improve productivity of many field operations by reducing operator fatigue and by improving machine performance. A self-tuning steering controller, which can be used for noncontact positioning systems, was designed for tractor guidance. A simple two degrees-of-freedom model was chosen with which to develop a prediction model used in recursive least-squares parameter estimation. A variable forgetting factor was implemented and its algorithm modified to cope with time-varying nonlinear systems. The self-tuning steering controller based upon minimum variance control was tested and verified with a tractor dynamic simulator. Circular, lane change, and sinusoidal paths were used. A model tractor was also used to verify controller performance for staight-line and sinusoidal paths.