A characteristic of wood chip refiners is that the static gain between the input (plate gap) and the output (motor load) is both nonlinear and time-varying, with reversal in the sign of the gain indicating the onset of pulp pad collapse towards lower values of the plate gap. The control objective is to regulate the motor load while avoiding pad collapse. The problem is principally stochastic in nature, since the gap at which gain reversal occurs can wander unpredictably. The control strategies developed in this paper are based primarily on viewing the refiner as a plant with uncertain linear characteristics. The proposed strategy consists of a multimodel parameter estimator (AFMM), a new active suboptimal dual controller, and some simple heuristic logic to deal with the nonlinearities. Several different combinations were evaluated by conducting a series of trials on an industrial refiner. The results show that while a more conventional estimator might have been used, the extra effort required to implement the dual controller appears to have been worthwhile, since probing aided significantly in the identification of gain changes and helped to prevent turn-off. The results also show that a variable retract strategy worked better than using a constant retract term for preventing operation in the pad collapse region, and that the control performance could be improved by identifying the maximum load the first time a pad collapse is encountered, and then using this information to set an upper limit on the set point.