I constructed a stochastic, spatially-explicit landscape model to seek optimal forest management decisions for long-term persistence of populations of red-cockaded woodpecker (Picoides borealis) and wood thrush (Hylocichla mustelina) on the Piedmont National Wildlife Refuge in Georgia, USA. I addressed uncertainty in decision making by considering alternative model forms that expressed different mechanisms of response by the forest and the bird populations to silvicultural actions. The implication of model uncertainty in this system is that conservation tradeoffs for both species differ according to choice of model. Decision variables in each model were the spatial scheduling of forest compartments for silvicultural treatments and the average periodicity of prescribed burning in compartments. Model responses were the number of active woodpecker clusters and abundance of wood thrushes. Additionally, I obtained a composite response as the average of the two abundance responses, each scaled by its standard error. I simulated each model under extremes of the decision alternatives, and I found a near-optimal management schedule for each model and for each of the responses. I also found near-optimal schedules for the case of complete uncertainty with regard to all models in the model set. Forest and bird monitoring data collected on the Refuge are the means by which measures of belief in each model are updated and decisions are adaptively improved. In nearly all models, both species responded strongly, but in opposite directions, to burning, and woodpeckers were sensitive to compartment scheduling. Consequently, optimal decisions were mostly similar among models, and values of information computed for each response suggested that little would be gained in management performance by resolving uncertainty among these models. However, fundamental uncertainties in the management of this system were probably not captured in this model set, and adaptive approaches therefore still hold promise for Refuge management. Current impediments to conducting adaptive management on the Refuge are (1) uncertainties regarding objectives, (2) lack of a comprehensive forest monitoring system, (3) inadequate system models, and (4) constraints in the expression and breadth of decision alternatives. I discuss critical information needed for the adaptive management of this and similar resource systems.