The standard predictor for the step size that might be used by a Runge-Kutta method is based on information gathered only in the current step. Previous work has shown that in some circumstances it is possible to make much better predictions using additional information from a previous step. Better predictions reduce the number of (expensive) failed steps. The author studies the new issues raised when solving stiff problems and develops a scheme using ideas from control theory that appears to be significantly better than the standard one.