Rangeland degradation within the arid zone of Western Australia has occurred as a consequence of sheep overstocking. Optimum grazing management strategies and rangeland rehabilitation techniques are needed to maintain the resource base for future use. In this paper an optimal control framework is developed for the derivation of grazing management decisions. " An integrated model of an arid grazing ecological system (IMAGES)" is used to derive the rangeland dynamics. The state of the grazing ecological system is summarized into four variables: the population of mature desirable perennial plants, young desirable perennial seedlings, old desirable perennial seedlings and total forage biomass. The controls are a set of different seasonal stocking rates within the year. Optimal decision rules are derived for both a deterministic and stochastic case study. Generally, the optimum stocking rate increase with increasing value for the 4 state variables. In deterministic case it combines both uniform and varied stocking rates while in the stochastic case only the uniform stocking rates prevails. Compared to the stochastic case, the net present value is higher for most cases under deterministic climate sequences, although there are some exceptions. The differences in the deterministic and stochastic cases can be dramatic. The reason appears to be highly variable rainfall combined with nonlinear production functions and adjustment costs. An over-optimistic expectation about the weather can be very expensive. Work to verify the stochastic results is continuing.