Environmental weeds are plants that invade natural ecosystems and present a serious threat to conservation of nature. Environmental weeds have been implicated in the extinction of several indigenous plant species, and they also threaten ecosystem stability and functional complexity. Historically, emphasis in weed control in Australian national parks has been placed on chemicals, manual pulling of small plants, excluding tourists and feral animal control measures. Recently, biological control has been introduced to control weed infestations. These methods typically have been applied as funds have become available, with little opportunity to consider their long-term effectiveness. As the threat from environmental weeds is becoming more fully recognised, an integrated, strategic, ecological and economic approach to weed management is needed. A deterministic dynamic programming model is developed for this purpose in this paper. A case study for scotch broom is presented, to assess the ways in which this approach can address the policy issues that face the community in the management of an environmental weed in a national park. The model takes account of the weed population dynamics, the effectiveness and cost of control measures, and the value of the park outputs (biodiversity, recreation and grazing). The dynamic programming model includes weed density and seed bank as state variables and a budget constraint for the control variables. The model is used to derive optimal control rules for any given state of the weed population. An optimal decision rule provides a package of control measures that can be used to attack the problem each year, depending on the current weed density and seed bank. Optimal trajectories are developed for a planning horizon of 45 years, and the effect of the budget constraint is analysed. The marginal value of an extra dollar for weed control, in terms of discounted future benefits, is estimated. It is shown that a combination of controls that targets both weed density and the seed bank is important. It is also shown (given the assumed parameters) that biological control is worth undertaking, as it appears as part of most of the optimal strategies identified. The paper concludes with policy recommendations for managing weeds in natural ecosystems.