Optimal any-aged management problems for mixed-species stands have been solved for the first time. Problem formulation calls for periodic planting and harvesting controls to be applied without constraints on the stand age or size structure over time; classical definitions of both even- and uneven-aged management are, thus, subsets of this general any-aged management definition. The solution technique is a derivative-free, coordinate-search process called the method of Hooke and Jeeves. The optimizer incorporates without modification the Stand Prognosis Model, a single-tree simulator that is used extensively in the western United States. This paper focuses on sensitivity analysis and performance of the optimizer on problems with both short and long time horizons and with different definitions of the harvest controls. Superior harvest regimes are found by defining harvest controls with relatively wide diameter-class boundaries and with broad species groups. In all cases examined, there is variability in present value and harvest pattern associated with the choice of the starting point for the optimizer. The cause of these multiple optima is a nonconvex and discontinuous response surface produced by the single-tree simulator. This points out the need to examine the results from several random starts before making conclusions about optimal timber harvesting.